Thomas Hazel

At the core of ChaosSearch #company #culture lies a passion for #pushing #boundaries - linkedin.com/feed/update/ur…

AI is Already Building AI: my conversation with @m__dehghani of @GoogleDeepMind about AI loops, recursive self improvement, continual learning and the latest in frontier AI. 00:00 Intro 01:17 What “loops” in AI actually mean 05:04 Recursive self-improvement as the next chapter of AI 07:32 @karpathy's autoresearch agents 08:56 AI building AI: how close are we? 10:02 The biggest bottlenecks: evals, automation, and long horizons 12:36 Can formal verification unlock recursive self-improvement? 14:06 What is model collapse? 15:33 Generalization vs specialization in AI 18:04 What is a specialized model today? 20:57 Could top AI researchers themselves be automated? 24:02 If AI builds AI, does data matter less than compute? 26:22 Post-training vs pre-training: where will progress come from? 28:14 Why pre-training is not dead 29:45 What is continual learning? 31:53 How real is continual learning today? 33:43 Mostafa’s background and path into AI 36:13 The story behind Universal Transformers 39:56 How Vision Transformers changed AI 43:47 Gemini, multimodality, and Nano Banana 47:46 Why multimodality helps build a world model 52:44 Why image generation is getting faster and more efficient 54:44 Hot takes section! 54:53 What the AI field is getting wrong 56:17 Why continual learning is underrated 57:26 Does RAG go away over time? 58:21 What people are too confident about in AI 59:56 What would you do if you were starting from scratch today?

@JeffDean Great!

RT @JeffDean: Today we're releasing Gemma 4, our new family of open foundation models, built on the same research and technology as our Gem…

𝐀𝐈 𝐚𝐠𝐞𝐧𝐭𝐬 𝐚𝐫𝐞 𝐛𝐞𝐜𝐨𝐦𝐢𝐧𝐠 𝐦𝐨𝐫𝐞 𝐚𝐧𝐝 𝐦𝐨𝐫𝐞 𝐢𝐦𝐩𝐨𝐫𝐭𝐚𝐧𝐭 𝐢𝐧 𝐨𝐮𝐫 𝐞𝐜𝐨𝐧𝐨𝐦𝐲 - linkedin.com/feed/update/ur…

The SaaS era was defined by unbundling : find a workflow, optimize it, own it. Salesforce chose sales automation. Slack chose chat. Dropbox chose file sharing. Point solutions won by perfecting single workflows. The playbook : own one pain point, expand from there. AI is moving faster than anyone predicted. When models change every 42 days, buyers can’t assemble a best-of-breed stack. They want a platform they can trust for three to five years.

Remembering Leonard on what would have been his 95th Birthday 🎂

At 95, I'm still smokin'! 😝 I’ve learned two things: Never waste a good cigar. Never trust anyone who says you should ‘act your age.’ 😉👍🏻

If you haven’t been paying attention, 𝐰𝐞’𝐫𝐞 𝐞𝐧𝐭𝐞𝐫𝐢𝐧𝐠 𝐚 𝐭𝐨𝐤𝐞𝐧 𝐞𝐜𝐨𝐧𝐨𝐦𝐲. The current model is straightforward: frontier models plus agent workloads mean more reasoning, more steps, and more tokens. As usage grows, costs grow with it - linkedin.com/feed/update/ur…

𝐅𝐫𝐨𝐦 𝐒𝐭𝐚𝐭𝐞𝐥𝐞𝐬𝐬 𝐀𝐈 𝐭𝐨 𝐂𝐨𝐦𝐩𝐨𝐮𝐧𝐝𝐢𝐧𝐠 𝐈𝐧𝐭𝐞𝐥𝐥𝐢𝐠𝐞𝐧𝐜𝐞 - AI in the enterprise is often framed as a tradeoff between cost and speed. The real equation is 𝐂𝐨𝐬𝐭 × 𝐒𝐩𝐞𝐞𝐝 × 𝐀𝐜𝐜𝐮𝐫𝐚𝐜𝐲 × 𝐒𝐜𝐚𝐥𝐞. - latentspin.ai/insights/cost-…

Databricks started later. It built a more complex architecture. It focused on unstructured data; images, documents, logs, audio. Though vast within the enterprise, this data had historically produced little insight. Too hard to process. Too messy to query. Too expensive to store in formats that mattered. Snowflake took the opposite bet. Structured data. Clean tables. SQL queries that ran fast & returned answers executives could read. The market agreed. Snowflake went public at a $70 billion valuation. Databricks raised private rounds at half that. Then AI arrived. Suddenly the data that was too messy to query became the data that models needed to train. Unstructured data wasn’t a liability. It was the asset. Databricks has overtaken Snowflake in revenue. Two years ago, Snowflake led by $220 million per quarter. Today, Databricks leads by $120 million. Databricks’ growth rate is accelerating at scale, from 50% to 55% to 65% year over year. Growth rates don’t accelerate at $5 billion in revenue. The crossover happened because AI is an architectural transition, not a feature addition. Most enterprise data never made it into Snowflake. It sat in object storage, unstructured, waiting. Databricks built tools to use it there. No migration required.

Peter Steinberger is joining OpenAI to drive the next generation of personal agents. He is a genius with a lot of amazing ideas about the future of very smart agents interacting with each other to do very useful things for people. We expect this will quickly become core to our product offerings. OpenClaw will live in a foundation as an open source project that OpenAI will continue to support. The future is going to be extremely multi-agent and it's important to us to support open source as part of that.

Thanks for having me on the @latentspacepod, @swyx and @FanaHOVA! I enjoyed our discussion! Site with summary: latent.space/p/jeffdean Video: youtu.be/F_1oDPWxpFQ?si…

A couple years ago I had a conversation with 𝐃𝐚𝐫𝐢𝐨 about the future of AI and specifically Large Language Models (LLMs) and their potential capabilities... linkedin.com/feed/update/ur…

𝐒𝐮𝐩𝐞𝐫 𝐁𝐨𝐰𝐥 𝐰𝐞𝐞𝐤𝐞𝐧𝐝 always gets me thinking about how we access and consume information. It got me thinking about why the 𝐅𝐮𝐭𝐮𝐫𝐞 𝐨𝐟 𝐀𝐈 should be about 𝐨𝐰𝐧𝐞𝐫𝐬𝐡𝐢𝐩, 𝐧𝐨𝐭 𝐚𝐜𝐜𝐞𝐬𝐬 - linkedin.com/feed/update/ur…

Are Large Language Models A Dead End Or Simply Incomplete? hubs.li/Q041SVl-0 Written by @ThomasHazel of @latentspin

State of LLMs 2026 - my conversation with @rasbt 01:05 - Are the days of Transformers numbered? 14:05 - World models: what they are and why people care 06:01 - Small “recursive” reasoning models (ARC, iterative refinement) 09:45 - What is a diffusion model (for text)? 13:24 - Are we seeing real architecture breakthroughs — or just polishing? 14:04 - MoE + “efficiency tweaks” that actually move the needle 17:26 - “Pre-training isn’t dead… it’s just boring” 18:03 - 2025’s headline shift: RLVR + GRPO (post-training for reasoning) 20:58 - Why RLHF is expensive (reward model + value model) 21:43 - Why GRPO makes RLVR cheaper and more scalable 24:54 - Process Reward Models (PRMs): why grading the steps is hard 28:20 - Can RLVR expand beyond math & coding? 30:27 - Why RL feels “finicky” at scale 32:34 - The practical “tips & tricks” that make GRPO more stable 35:29 - The meta-lesson of 2025: progress = lots of small improvements 38:41 - “Benchmaxxing”: why benchmarks are getting less trustworthy 43:10 - The other big lever: inference-time scaling 47:36 - Tool use: reducing hallucinations by calling external tools 49:57 - The “private data edge” + in-house model training 55:14 - Continual learning: why it’s hard (and why it’s not 2026) 59:28 - How Sebastian works: reading, coding, learning “from scratch” 01:04:55 - LLM burnout + how he uses models (without replacing himself)

🙄… thoughts @PattiMurin‽

2.67 years of AI progress

AI Coworkers (i.e. Agents) that learn is essential - linkedin.com/feed/update/ur…

Will agents need Free Will to execute complex tasks? linkedin.com/feed/update/ur…