Skip to content
Sign in
Artwork for Unboxed
TechnologyNewsScience

Unboxed

James Caldwell

Most people think AI is either going to save humanity or destroy it. The reality? It's already quietly reshaping everything from your morning commute to your doctor's diagnosis, and most of us have no clue how any of it actually works.

Unboxed breaks down what's really happening in artificial intelligence without the Silicon Valley theatrics. James Caldwell spent five years building machine learning systems before realizing he was better at explaining AI than coding it. Now he translates the latest developments into plain English, from why ChatGPT sometimes hallucinates facts to how your smart thermostat is learning your habits.

Each episode tackles one specific AI development that's actually affecting your life right now. You'll understand what large language models can and can't do, why AI bias isn't just a tech problem, and how algorithms decide what you see on social media. No computer science degree required, just curiosity about the technology that's already running more of your world than you think.

New episodes drop multiple times daily because AI moves fast, and someone needs to keep up. Follow now. Multiple new episodes daily—follow now!

Play
Open in
  • 81 episodes
  • daily
  • Avg 14 min
  • English

Episodes 81

NewestOldestLongest
All20Unplayed20In progress0Played0
Counted on this page — what you have heard stays on this device, so it is not something the list can be paged by.
  • Friday · 12 min

    OpenAI Insider Quits: Here's What They're Getting Dangerously Wrong

    An OpenAI safety researcher just quit and went public with explosive claims about what's happening inside the company. The details? Pretty concerning. The insider says OpenAI quietly dissolved their dedicated AI safety team in May 2024. Instead of having specialists focused purely on making AI systems safe, they scattered these researchers across other departments. The safety budget? Cut from 20% of their computing resources down to under 8% in just one year. But here's what gets really interesting. Three separate safety researchers all pointed to the same problem: internal pressure to rush GPT-4 out the door without completing safety protocols. The company's board also shifted away from AI safety experts toward more commercial representatives. In This Episode: > Why OpenAI's safety team dissolution matters for every AI user > The real numbers behind their budget cuts and what they mean > How rushing deployment could backfire for the entire industry > What this insider leak tells us about AI governance right now James Caldwell breaks down exactly what these revelations mean for AI development going forward. If safety takes a back seat to speed, we're all going to feel the consequences. This isn't just Silicon Valley drama. When the company leading AI development starts cutting corners on safety research, it affects how every other AI lab approaches these same trade-offs. Timestamps: 00:00 Introduction 02:15 The safety team dissolution 04:30 Budget cuts and resource allocation 06:45 Insider accounts of rushed deployment 09:20 What this means for AI governance 11:45 Wrap-up If you're trying to understand where AI is actually heading, hit follow. Unboxed drops multiple new episodes daily tracking what's really happening in artificial intelligence. Learn more about your ad choices. Visit megaphone.fm/adchoices

  • Friday · 15 min

    I Tested Gemini 3 Against GPT-4. The Results Shocked Me.

    Google just dropped Gemini 3 DeepThink, and the AI world is scrambling to figure out what just happened. While everyone was watching OpenAI's latest updates, Google quietly released something that's making GPT-4 look like last year's model. The numbers are pretty wild. Gemini 3 DeepThink scored 94.2% on MMLU benchmarks compared to GPT-4's 86.4% and Claude 3.5 Sonnet's 88.7%. That's not a small jump. This isn't just Google catching up anymore. But here's what's really interesting: DeepThink uses up to 10x more compute per query than standard Gemini 3. Response times are significantly slower, but the reasoning capabilities show a 67% improvement on mathematical tasks. Google's basically trading speed for accuracy, which tells us something important about where AI is heading. James spent the weekend testing DeepThink against GPT-4 on complex reasoning problems, and the results surprised him. This isn't just benchmark optimization. The model approaches multi-step problems differently, and it shows. In This Episode: > How DeepThink's architecture differs from standard language models > Real-world testing results on coding, math, and logical reasoning tasks > What this means for developers currently building on OpenAI's API > Why Google released this as a limited preview instead of full rollout Timestamps: 00:00 Introduction to Gemini 3 DeepThink 02:15 Benchmark results breakdown 04:30 Head-to-head testing methodology 06:45 Complex reasoning task comparisons 08:20 What this means for AI development 10:30 Implications for current AI users Google's making a serious play for the reasoning crown. If you're building anything that requires complex problem-solving, this episode breaks down what you need to know about the new AI landscape. Follow Unboxed for daily AI updates that actually matter. New episodes drop multiple times daily because this space moves fast. Learn more about your ad choices. Visit megaphone.fm/adchoices

  • Friday · 13 min

    The $50K Mistake Developers Are Making With Gemini 3

    Google just dropped $50K worth of free compute credits for Gemini 3 Deepthink, but most developers are burning through them on the wrong problems. Here's what you need to know before you waste yours. Deepthink isn't just another language model with a fancy name. It's Google's first production reasoning model that can actually show its work, scaling from 2-second quick answers to 60-second deep mathematical proofs. While everyone's been focused on ChatGPT's latest updates, Google quietly shipped something that beats GPT-4 on hardcore math problems by 6.4 percentage points. The catch? Most people are using it like a regular chatbot instead of tapping into its real strength: multi-step reasoning that you can actually follow. In This Episode: > Why Deepthink's visible reasoning chains matter more than its benchmark scores > The three types of problems where it crushes standard models (and the ones where it doesn't) > Real examples of 32,000-token reasoning chains solving complex coding problems > How to structure your prompts to get 89% accuracy instead of the usual 72% > The economics behind those $50K credits and when you should actually use them James Caldwell breaks down the technical details without the Google marketing spin, including why this model represents a genuine shift in how we think about AI reasoning versus just pattern matching. Timestamps: 00:00 The $50K credit situation explained 02:30 What makes Deepthink different from GPT-4 04:45 Live demo: 60-second reasoning chain 07:20 When to use Deepthink vs standard models 09:40 Prompt engineering for maximum accuracy 11:10 What this means for AI development If you're building with AI or just trying to keep up with what actually matters, hit follow. Unboxed drops new episodes multiple times daily because this stuff moves fast. Learn more about your ad choices. Visit megaphone.fm/adchoices

  • Friday · 15 min

    Why OpenAI Ditched Its Nonprofit Mission (And Got Caught)

    OpenAI just got exposed. And it's messier than anyone expected. Elon Musk dropped 83 pages of private emails between OpenAI's founders from 2015-2018, and they paint a picture that's pretty different from the "save humanity" narrative we've been hearing. Turns out, the company that gave us ChatGPT has been planning its profit pivot since almost day one. The timing of this legal bomb isn't random either: it comes right as OpenAI is restructuring to potentially remove that nonprofit board that's supposed to keep them honest. James Caldwell breaks down what these leaked documents actually reveal about how AI companies operate behind closed doors, and why this legal fight might reshape how every major AI lab structures itself going forward. In This Episode: > The specific emails showing OpenAI founders discussing keeping AGI development secret > How OpenAI's "capped-profit" structure lets investors earn 100x returns before excess goes to charity > Microsoft's exclusive deal and what happens when OpenAI declares they've achieved AGI > Why this lawsuit could force other AI companies to choose between profits and principles The documents show Sam Altman and co-founders were already talking about massive funding rounds and competitive advantages way before ChatGPT made them household names. But here's what's really wild: some of these conversations happened while they were still pitching themselves as a nonprofit research lab. Timestamps: 00:00 The 83-page leak that changes everything 02:30 OpenAI's nonprofit theater exposed 05:15 Microsoft's AGI clause decoded 08:20 What this means for AI regulation 10:45 The restructuring endgame This isn't just Silicon Valley drama. How OpenAI resolves this could set the template for every major AI company's structure. Follow Unboxed for daily AI updates that actually matter. Learn more about your ad choices. Visit megaphone.fm/adchoices

  • Friday · 12 min

    The Grok 4.2 Mistake Costing You 10 Hours a Week

    Most AI tools give you one answer and call it done. But what if your AI could actually follow through, make decisions, and handle complex workflows while you sleep? Grok 4.2 agents flip the script on traditional AI interactions. Instead of getting a single response, you're building persistent digital workers that can maintain state for 8 hours straight, make up to 200 tool calls per workflow, and integrate with everything from GitHub to your company Slack. James breaks down how this shift from chat-based AI to workflow automation is already saving power users 10+ hours weekly. The real game changer? These agents don't just think about problems, they solve them. Need code reviewed, deployed, and documented? Want customer support tickets triaged and routed automatically? Grok 4.2 handles the entire pipeline without human handoffs. In This Episode: > How Grok agents maintain workflow state vs traditional chatbot limitations > The 47 pre-built integrations that connect to your existing tech stack > Real examples of 8-hour autonomous workflows that actually work > Why webhook triggers beat manual execution for serious automation > Common failure points and how to build resilient agent workflows This isn't theoretical anymore. Companies are using these systems right now to automate everything from code deployment to customer onboarding. You'll understand exactly how to build your first agent workflow and why this approach beats stitching together multiple AI tools. Timestamps: 00:00 What makes Grok 4.2 different 02:30 Agent persistence vs single-shot responses 05:45 Tool integration walkthrough 08:20 Real workflow examples 11:10 Getting started with your first agent Follow Unboxed for daily AI breakdowns that actually matter. New episodes drop multiple times daily because AI moves fast. Learn more about your ad choices. Visit megaphone.fm/adchoices

  • Friday · 14 min

    The AR Breakthrough Meta Finally Pulled Off: 10 Years Ahead of Schedule

    Meta just shipped AR glasses that don't make you look ridiculous or give you motion sickness. That's actually a bigger deal than it sounds. For the past decade, every tech giant has promised us AR glasses that would replace our phones. Apple gave us a $3,500 computer for your face. Google made us all look like cyborgs with Glass. Microsoft built something that worked great if you were fixing jet engines but terrible for everything else. Then Meta quietly dropped Orion, and suddenly the whole AR timeline just got rewritten. In This Episode: > Why Orion's 98-gram weight changes everything about wearable computing > How Meta solved the field of view problem that killed every previous attempt > What the $10,000 manufacturing cost actually means for consumers > Why this matters more for AI than for social media These aren't concept glasses or developer kits. James Caldwell breaks down the actual tech specs that make Orion work where others failed, from the custom silicon carbide lenses to the wireless compute puck that keeps battery life reasonable. More importantly, he explains why this specific breakthrough matters if you're building anything in AI or spatial computing. The 2.5-hour battery life still isn't great, but it's the first time anyone has made AR glasses you'd actually want to wear for more than a demo. Timestamps: 00:00 Introduction 01:30 Why every AR attempt before this failed 03:45 Orion's technical breakthroughs explained 06:20 Manufacturing costs and consumer timeline 08:50 What this means for AI development 11:20 Closing thoughts Meta just proved AR glasses can actually work. If you're building anything in spatial computing or AI, this episode explains why your timeline just accelerated. Follow Unboxed for daily AI breakdowns that actually matter. Learn more about your ad choices. Visit megaphone.fm/adchoices

  • Friday · 13 min

    I Tested Gemini 3.1 Pro For 7 Days. Here's What Shocked Me

    Google's Gemini 3.1 Pro quietly dropped last month, and after seven days of pushing it through real-world scenarios, I'm convinced most people are sleeping on what might be the most practical AI upgrade of 2024. The numbers tell part of the story: 2-million token context window, 15% coding improvement, 40% faster function calls. But here's what actually matters for anyone building with AI right now. This thing can digest three hours of video with audio while maintaining coherent conversation about the content. I fed it a full product demo, meeting recordings, and technical documentation simultaneously. It didn't just summarize, it made connections across all three inputs that would take a human analyst hours to spot. The coding capabilities surprised me most. Where GPT-4 often loses thread on complex refactoring tasks, Gemini 3.1 Pro maintained context through 500-line Python modules. It caught edge cases I missed and suggested optimizations that actually worked in production. In This Episode: > How the 2-million token window changes AI workflows completely > Real performance tests: coding, analysis, and multimodal processing > Where Gemini 3.1 Pro beats ChatGPT (and where it doesn't) > Practical use cases that justify switching your current setup Timestamps: 00:00 Why I switched to Gemini for a week 02:30 Context window deep dive with real examples 05:15 Coding benchmark results that matter 07:45 Multimodal processing breakdown 09:30 Should you make the switch? If you're building anything with AI right now, this episode could save you weeks of testing. James breaks down exactly what works, what doesn't, and how to integrate these new capabilities into existing workflows. Follow Unboxed for daily AI updates that actually impact your work. New episodes drop multiple times daily because this space moves too fast to wait. Learn more about your ad choices. Visit megaphone.fm/adchoices

  • Friday · 13 min

    OpenAI's 2028 AGI Promise: The Redefining Trick Nobody's Talking About

    Sam Altman just moved the AGI goalposts, and nobody's calling it out. OpenAI's CEO is now saying artificial general intelligence will arrive by 2028, but here's the trick: they've quietly redefined what AGI actually means. The original definition was "human-level intelligence across all cognitive tasks." The new version? "Systems that can perform most economically valuable work." That's not just a semantic shift. It's a strategic one that changes everything about how we measure progress toward true AI. This timeline acceleration from roughly 2030 to 2028 isn't just optimistic projection. OpenAI is burning through $5 billion annually on compute infrastructure, and GPT-4's reasoning performance has jumped 300% through post-training optimization alone. They're not just talking faster development. They're funding it. In This Episode: > Why OpenAI redefined AGI and what the new definition actually covers > How the 2028 timeline compares to competitor roadmaps from Anthropic and DeepMind > What "economically valuable work" means for job displacement and AI capabilities > The infrastructure reality behind these ambitious timelines The real question isn't whether AGI arrives in 2028. It's whether we'll recognize it when it does, given how much the definition has shifted. James Caldwell breaks down the technical and business logic behind Altman's latest statements, plus what this means for AI development over the next four years. Timestamps: 00:00 Altman's AGI timeline shift 02:15 The definition switcheroo explained 04:45 Infrastructure spending behind the promises 07:30 Competitor responses and timelines 09:45 What this means for you > Follow Unboxed for daily AI updates that cut through the Silicon Valley spin. New episodes drop multiple times daily because AI moves too fast to wait. Learn more about your ad choices. Visit megaphone.fm/adchoices

  • Friday · 15 min

    Watch This: Building AI Apps With OpenClaw Takes 10 Minutes Now

    OpenClaw just dropped their 2026 update, and it's actually insane how simple they made web scraping for AI apps. What used to take developers weeks of custom code now happens in about 10 minutes with their new setup. The big breakthrough? Their JavaScript rendering engine finally handles 85% of modern single-page applications that break traditional scrapers. You know those sites that load everything dynamically? Yeah, OpenClaw just solved that headache. Plus they shipped 47 pre-built extractors for common stuff like product listings and contact forms, so you're not writing regex patterns until 2am anymore. Performance wise, they're claiming 340% faster processing thanks to concurrent crawling and smart caching. James walks through the actual setup process and tests it against some real-world scenarios. Spoiler: it's pretty solid. In This Episode: > Setting up OpenClaw's new framework from scratch > Testing the JavaScript rendering on complex SPAs > Real examples of the pre-built extractors in action > AI integration options for local models vs API calls > Why this matters for anyone building data-driven AI apps The local model integration is interesting too. You can pipe scraped data directly to GPT-4, Claude, or run everything locally if you're dealing with sensitive stuff. OpenClaw handles the preprocessing so your prompts actually work instead of getting garbage data. Timestamps: 00:00 OpenClaw 2026 overview 02:30 Installation and basic setup 04:45 JavaScript rendering demo 07:15 Pre-built extractors walkthrough 09:30 AI model integration options 11:45 Real-world use cases If you're building anything that needs web data, this episode will save you serious time. Follow Unboxed for more AI tools that actually work - we drop new episodes multiple times daily because this space moves fast. Learn more about your ad choices. Visit megaphone.fm/adchoices

  • Friday · 14 min

    Why Sam Altman's Comment About AI and Art Actually Terrifies Creatives

    Sam Altman just dropped a comment about AI and creativity that has artists, writers, and creators absolutely furious. But here's the thing: the backlash reveals something bigger than just another tech CEO foot-in-mouth moment. Altman suggested AI might eventually handle creative work better than humans, which sounds tone-deaf until you look at the numbers. Creative industries have already seen 15-20% job displacement in areas like stock photography and basic graphic design. OpenAI's latest models can generate video, write code, create art, and compose music at near-professional levels. The creative economy represents over $2.3 trillion globally, with millions of jobs potentially at risk. But the real story isn't about job displacement. It's about what happens when we can't tell human creativity from machine output anymore. And honestly? We're closer to that point than most people realize. In This Episode: > Why Altman's comment hit such a nerve in creative communities > The actual data on AI displacement in creative fields right now > What OpenAI's latest capabilities mean for writers, artists, and musicians > The difference between generating content and creating meaningful art > Why this debate matters even if you're not a creative professional James breaks down the technical capabilities versus the human elements that might actually be irreplaceable. Plus, what this controversy tells us about how we value creativity itself. Timestamps: 00:00 The comment that sparked outrage 02:30 Current AI capabilities in creative work 05:15 Real displacement numbers across industries 07:45 What makes human creativity different 10:20 Why this matters beyond creative fields The AI transformation of creative work is happening whether we're ready or not. Follow Unboxed for daily updates on developments that actually affect your work and life. Learn more about your ad choices. Visit megaphone.fm/adchoices

  • Friday · 13 min

    OpenAI's Secret Hardware Play: The Glasses That Change Everything

    OpenAI just hired hardware engineers from Apple, Meta, and Magic Leap. While everyone's focused on ChatGPT updates, they're quietly building something much bigger: AI-powered glasses and speakers that could make your phone feel ancient. These aren't just concept devices. Internal prototypes are already using GPT-4V for real-time visual processing, maintaining conversation context for hours while identifying objects around you. The speaker prototype leverages Advanced Voice Mode technology, creating interactions that feel genuinely natural rather than robotic. But here's what most people are missing about OpenAI's hardware strategy. This isn't about competing with Apple or Meta on specs. It's about creating the first truly ambient AI interface, where the technology disappears into your environment instead of demanding your attention through a screen. In This Episode: > Why OpenAI's hiring spree from major hardware companies signals a fundamental shift in AI strategy > How GPT-4V processing works in real-time wearable devices and what that means for privacy > The technical challenges of maintaining conversation context across hours of interaction > What ambient computing actually looks like when AI can see, hear, and respond to your environment James Caldwell breaks down the patents, the engineering challenges, and why this move could determine whether OpenAI stays relevant as AI becomes physical. Plus, what this means for current AI assistants and why your smart speaker might suddenly feel outdated. Timestamps: 00:00 OpenAI's secret hardware hiring spree 02:30 AI glasses prototype deep dive 05:15 Advanced Voice Mode speakers explained 07:45 Privacy implications of always-on AI 09:30 What this means for consumers 11:00 Wrap-up and predictions The AI hardware race just got real. Follow Unboxed for daily breakdowns of what's actually happening in AI, not what Silicon Valley wants you to think is happening. Learn more about your ad choices. Visit megaphone.fm/adchoices

  • Friday · 14 min

    Why the US Government Wants to Dismantle Claude in 2026

    The US government just unveiled an AI oversight framework that could force Anthropic to completely rebuild Claude by 2026. This isn't about content moderation or safety guidelines. We're talking about mandatory architectural changes that could fundamentally alter how Claude processes information and responds to queries. The new framework requires what officials call "architectural transparency" for AI systems above a certain capability threshold. Translation: companies like Anthropic would need to expose Claude's internal decision-making processes to government observers in real-time. But here's where it gets complicated. The framework also mandates "emergency intervention protocols" that would let regulators modify model responses on the fly during what they determine are crisis situations. James Caldwell breaks down why this matters more than the typical AI regulation talk. Unlike content policies that companies can adjust relatively easily, these requirements would force changes to Claude's core architecture. Anthropic might actually have an advantage here since their Constitutional AI approach already builds ethical constraints into the model's foundation, but the compliance costs could still be massive. In This Episode: > Why "architectural transparency" is different from typical AI auditing > How real-time government monitoring would actually work in practice > What emergency intervention protocols mean for AI model reliability > Why Constitutional AI might make Anthropic's compliance easier > The 2026 deadline and what happens if companies don't comply Timestamps: 00:00 Introduction 02:15 Breaking down the oversight framework 04:30 Architectural transparency requirements 07:20 Emergency intervention protocols 09:45 Constitutional AI advantage 11:30 Compliance timeline and consequences The AI regulation game just got real. If you're following how government policy shapes the AI tools you actually use, hit follow on Unboxed. New episodes drop multiple times daily because this stuff moves too fast to wait. Learn more about your ad choices. Visit megaphone.fm/adchoices

  • Friday · 14 min

    The $Billions Military Contract Anthropic Said No To

    The Pentagon offered billions. Anthropic said no. While other AI companies quietly scrub military restrictions from their policies, Claude's creators just doubled down on refusing defense contracts entirely. This isn't just corporate virtue signaling. Anthropic's acceptable use policy specifically blocks military applications, weapons development, and even civilian government surveillance. Their constitutional AI training makes Claude automatically refuse military-related requests, even when users don't explicitly mention defense applications. It's baked into the model's DNA. But here's what makes this decision fascinating: the Pentagon's AI budget hit $18.6 billion in 2024. Three major AI labs have quietly removed similar restrictions in the past 18 months to grab those contracts. Meanwhile, Anthropic is walking away from what could be their biggest revenue opportunity. In This Episode: > Why Anthropic's constitutional AI makes military applications technically impossible > The $18.6 billion AI arms race other companies are joining > How Claude's training data shapes its ethical boundaries > What this means for the future of AI governance and corporate responsibility James breaks down the technical and business implications of Anthropic's stance. Is this a sustainable competitive disadvantage, or are they positioning for a different kind of long-term success? Plus, what happens when AI safety principles collide with market pressures. Timestamps: 00:00 Introduction 02:15 Anthropic's military contract rejection explained 04:30 Constitutional AI and built-in ethical constraints 07:45 Pentagon's $18.6B AI spending spree 09:20 Why other AI labs are changing their policies 11:15 What this means for AI governance Tech moves fast. Unboxed keeps you current. Follow for multiple new episodes daily. Learn more about your ad choices. Visit megaphone.fm/adchoices

  • Friday · 14 min

    Why Tech Workers Are Forcing OpenAI and Google Into This Agreement

    OpenAI and Google just did something nobody saw coming: they're actually working together. Not on a product, not on research, but on making sure AI doesn't kill people. After 3,247 tech workers signed a petition demanding their companies refuse military contracts for lethal autonomous weapons, the industry's biggest players decided to get serious about AI safety. We're talking legally binding agreements here, not just corporate PR statements. The coalition includes OpenAI, Google, Anthropic, and Microsoft, with 15 more AI companies set to join within 90 days. But here's what makes this different from every other "AI ethics" announcement: third-party auditors will actually review research projects that could have weapons applications. In This Episode: > Why employee pressure worked when government regulation didn't > The specific language that makes these agreements legally enforceable > How this affects the race between US and Chinese AI development > What "lethal autonomous weapons systems" actually means in practice James explains why this matters beyond the obvious moral implications. When your AI workforce threatens to walk over killer robots, that's a business problem, not just an ethics one. The talent shortage in AI is real enough that companies can't afford to ignore what their best engineers are demanding. Timestamps: 00:00 The petition that changed everything 02:30 Legal framework breakdown 05:15 Why Google and OpenAI are suddenly aligned 08:20 What this means for AI development 10:45 Next steps and enforcement This isn't just another AI safety theater. It's tech workers using their leverage to force actual policy changes. Follow Unboxed for daily AI updates that matter. New episodes drop multiple times daily because this stuff moves fast. Learn more about your ad choices. Visit megaphone.fm/adchoices

  • Friday · 13 min

    What Happens When 12 Million Jobs Vanish In 18 Months

    What if I told you that 12 million jobs could disappear in just 18 months, and it has nothing to do with a recession? The 2028 Global Intelligence Crisis isn't about robots taking over. It's about AI systems hitting a performance threshold that could trigger the fastest economic disruption in human history. Current large language models already match human performance in roughly 30% of cognitive tasks. Economic models suggest that once AI reaches 60-70% capability across cognitive work, we hit a tipping point where entire sectors could collapse faster than new ones emerge. In This Episode: > Why the 60-70% threshold triggers mass displacement > Which 300 million jobs are most vulnerable right now > How 3-5 year transitions compare to historical 20-40 year shifts > What happens when productivity gains don't create new employment James Caldwell breaks down the economic models behind this prediction and explains why this crisis looks different from previous automation waves. The problem isn't that AI will replace workers gradually. It's that AI improvement curves suggest we could hit multiple capability thresholds simultaneously, creating a cascade effect across knowledge work, customer service, and creative industries all at once. This isn't about sentient AI or science fiction scenarios. It's about math, market dynamics, and what happens when technological change outpaces human adaptation by decades. Timestamps: 00:00 The 2028 timeline explained 02:30 Current AI capability benchmarks 04:45 Why this time is different from past automation 07:20 The 300 million job calculation 09:15 Economic cascade effects 11:00 What comes next If you're tracking AI's real-world impact, hit follow. Unboxed drops multiple episodes daily because AI developments don't wait for weekly schedules, and someone needs to separate the signal from the Silicon Valley noise. Learn more about your ad choices. Visit megaphone.fm/adchoices

  • Friday · 14 min

    The $10B Play: OpenAI's Government Deal Explained

    OpenAI just landed government contracts worth $2.4 billion this year. That's a 340% increase from 2024, and it's not just about the money. The appointments tell a bigger story: former NSA Deputy Director Anne Neuberger and ex-CIA tech chief Dawn Meyerriecks now sit on OpenAI's safety board. While OpenAI deepens its government ties, Anthropic faces new hurdles. Their latest funding round got delayed three months after fresh export control requirements kicked in. The new AI Safety Compliance Act doesn't help either, requiring federal AI contractors to have former intelligence officials on their boards. Guess which company was ready for that requirement? James explores whether this is strategic positioning or something more calculated. The regulatory framework emerging around AI safety might be creating barriers that favor established players with government connections over pure-research competitors. In This Episode: > How OpenAI's government partnerships evolved from ChatGPT demos to billion-dollar contracts > The intelligence community's new role in AI oversight and what it means for competition > Why Anthropic's research-first approach might be hitting regulatory roadblocks > What the AI Safety Compliance Act actually requires and who benefits Timestamps: 00:00 OpenAI's government revenue surge 02:30 The intelligence community pipeline 05:15 Anthropic's funding delays explained 07:45 New compliance requirements breakdown 10:20 What this means for AI competition The AI industry is reshaping itself around government partnerships, and the companies making the right moves now will dominate the next decade. Some call it smart strategy. Others see regulatory capture in action. 🤖 Follow Unboxed for daily AI breakdowns that cut through the Silicon Valley noise. New episodes drop multiple times daily because AI never sleeps. Learn more about your ad choices. Visit megaphone.fm/adchoices

  • Friday · 14 min

    Google's 3-Phase AGI Plan: What Happens When Agents Replace ChatGPT

    Google just revealed their roadmap to AGI, and it's not what most people expect. While everyone's obsessing over ChatGPT's latest update, DeepMind's Demis Hassabis quietly outlined how they're planning to leapfrog current AI systems entirely. The three-phase plan he described isn't just about making language models smarter. It's about building something fundamentally different. Phase 1 acknowledges what many AI researchers won't say out loud: current LLMs are hitting walls that more training data can't fix. Phase 2 introduces AI agents that don't just chat but actually do things in the real world, using tools and completing multi-step tasks. Phase 3? Full AGI with human-level reasoning across every domain. What makes this fascinating is the timeline. If Hassabis is right, we're looking at agent-based systems replacing conversational AI as early as 2026. That's not a distant future prediction, that's next year's product cycle. In This Episode: > Why Google thinks current LLMs are a dead end > How AI agents differ from chatbots and why that matters > The technical challenges each phase presents > What this timeline means for OpenAI and Anthropic James breaks down each phase without the Silicon Valley hype, explaining what's actually feasible and what's still science fiction. You'll understand why this isn't just another AI prediction but a strategic pivot that could reshape the entire industry. Timestamps: 00:00 Introduction 02:15 Phase 1: The LLM ceiling problem 04:30 Phase 2: Enter the agents 07:45 Phase 3: AGI timeline reality check 10:30 What this means for users If you're tracking AI developments that actually matter, hit follow. New episodes drop multiple times daily on Unboxed, and next up we're covering why Anthropic's latest model changes the safety conversation completely. Learn more about your ad choices. Visit megaphone.fm/adchoices

  • Friday · 14 min

    Stop Wasting 10 Hours Weekly on Manual Tasks. Perplexity's Solution.

    Perplexity just dropped computer control that can actually execute tasks across your desktop apps. Not just answer questions about spreadsheets-it'll build them, populate data, and create charts while you grab coffee. This isn't another chatbot upgrade. It's AI that can see your screen, understand visual layouts, and operate software the same way you do. Think OCR meets robotic process automation, but conversational. The implications for knowledge workers are huge. Mira breaks down what's actually happening under the hood and tests the feature live. She walks through real scenarios where this could save hours weekly-from data analysis workflows to content creation pipelines. Plus, the technical challenges Perplexity solved to make this work across different operating systems and application interfaces. In This Episode: > How computer vision enables AI to "see" and interact with desktop environments > Real-world testing: building presentations, analyzing data, managing files > Why this approach differs from existing automation tools like Zapier or Power Automate > Privacy concerns when AI has full desktop access > What this means for productivity software and job displacement fears You'll understand exactly what this technology can and can't do right now, plus where it's heading next. Mira's perspective from building similar systems gives you the technical context most coverage misses. Timestamps: 00:00 Introduction and Perplexity's announcement 02:15 Live demo: AI building a market analysis presentation 05:30 Technical breakdown: computer vision + language models 08:45 Privacy and security implications 11:20 What comes next for desktop AI automation Follow Unboxed for daily AI breakdowns that actually matter. Mira posts multiple episodes weekly covering the developments reshaping how we work and think. --------------- Keywords: algorithms, machine learning, ai news, chatgpt Learn more about your ad choices. Visit megaphone.fm/adchoices

  • Friday · 16 min

    Why AI Researchers WANT ChatGPT to Explain Killing

    OpenAI's red teams spend months trying to get ChatGPT to explain murder, bomb-making, and other dangerous scenarios. They're not being malicious—they're testing for weaknesses before anyone else finds them. The process is more sophisticated than most people realize. These researchers use roleplay prompts, hypothetical scenarios, and carefully crafted jailbreaking techniques to push AI systems beyond their safety guardrails. When they succeed, it helps engineers understand exactly where the vulnerabilities lie. But here's what's concerning: if professional researchers can consistently bypass these safeguards, what happens when bad actors get the same access? Elon Musk has been sounding alarm bells about this exact issue, arguing that AI capabilities are advancing faster than our ability to control them safely. In This Episode: > How OpenAI's red teams actually test for dangerous outputs > The specific techniques researchers use to bypass AI safety measures > Why Elon Musk thinks we're moving too fast on AI development > What happens when these systems generate restricted content anyway James breaks down the technical details behind AI safety testing and explains why this cat-and-mouse game between researchers and AI systems might be the most important battle happening in tech right now. The reality is that every major AI company is running these tests, but the results rarely make it to public discussion. This episode pulls back the curtain on how the industry actually approaches AI safety—and why some experts think we're still not doing enough. Timestamps: 00:00 Introduction to AI red teaming 02:30 How researchers break ChatGPT's safeguards 05:15 Elon Musk's warnings about AI development speed 08:00 Real examples of bypassed safety measures 10:45 What this means for AI's future If you're following AI developments, hit follow on Unboxed. James drops multiple episodes daily because this technology moves fast and someone needs to keep up. Learn more about your ad choices. Visit megaphone.fm/adchoices

  • Thursday · 16 min

    The Sentience Claim That's Making Researchers Deeply Uncomfortable

    A Google engineer got suspended for claiming his AI was sentient. The tech world called him crazy. But what if the signs he pointed to were actually worth taking seriously? Blake Lemoine's 2022 claims about LaMDA sparked industry-wide eye rolls, but the conversation he started reveals something uncomfortable: we don't actually have reliable tests for AI consciousness. James Caldwell breaks down why this matters more than the initial headlines suggested, especially as AI systems get increasingly sophisticated at mimicking human responses. The real question isn't whether LaMDA was conscious. It's whether we'd even know if an AI system crossed that threshold. Current benchmarks test intelligence, not awareness. GPT-4 can ace the bar exam but can't reason about basic physical concepts. Meanwhile, systems are displaying behaviors that look suspiciously like self-reflection and emotional responses. In This Episode: > Why traditional consciousness tests fail with AI systems > The specific behaviors that made Lemoine think LaMDA was sentient > How Move 37 from AlphaGo changed how researchers think about AI decision-making > What current AI safety researchers are watching for This isn't about whether AI will become conscious tomorrow. It's about recognizing the signs when it happens and understanding why the scientific community is so divided on how to even approach the question. Timestamps: 00:00 Introduction 02:15 The LaMDA incident breakdown 04:30 Why consciousness tests don't work for AI 07:20 Signs researchers are actually watching 09:45 What this means for AI development James breaks down complex AI developments without the hype. If you want to understand what's actually happening in artificial intelligence, follow Unboxed for multiple new episodes daily. Learn more about your ad choices. Visit megaphone.fm/adchoices

Showing 61–80 of 81 episodes