
Sign up to save your podcasts
Or


Hear me now, quote me later: these are the 8 ways in which ChatGPT will be improved in ChatGPT 4. From critical reasoning to coding, comprehension to calculation, I will give the peer-reviewed evidence of what is to come.
By examining publicly accessible benchmarks, comparable large language models and the latest research papers, we can discern the ways in which GPT4 (integrated into Bing or otherwise) will beat ChatGPT.
I'll show you how unreleased models already beat current ChatGPT and all of this will actually give a clearer insight into what even GPT5 and rival models from Google might well be able to achieve.
https://ai.googleblog.com/2022/04/pathways-language-model-palm-scaling-to.html
https://cloud.google.com/blog/topics/tpus/google-showcases-cloud-tpu-v4-pods-for-large-model-training
https://arxiv.org/pdf/2204.02311.pdf
https://arxiv.org/pdf/1905.00537.pdf
https://arxiv.org/pdf/2201.11903.pdf
https://github.com/google/BIG-bench/
https://github.com/google/BIG-bench/blob/main/bigbench/benchmark_tasks/README.md
https://github.com/google/BIG-bench/tree/main/bigbench/benchmark_tasks/gre_reading_comprehension
https://github.com/google/BIG-bench/tree/main/bigbench/benchmark_tasks/logical_args
https://github.com/google/BIG-bench/tree/main/bigbench/benchmark_tasks/evaluating_information_essentiality
https://github.com/google/BIG-bench/tree/main/bigbench/benchmark_tasks/physics
http://web.mit.edu/~yczeng/Public/WORKBOOK%201%20FULL.pdf
https://github.com/google/BIG-bench/tree/main/bigbench/benchmark_tasks/sufficient_information
https://github.com/google/BIG-bench/tree/main/bigbench/benchmark_tasks/implicatures
https://github.com/google/BIG-bench/tree/main/bigbench/benchmark_tasks/winowhy
https://www.deepmind.com/publications/an-empirical-analysis-of-compute-optimal-large-language-model-training
https://lambdalabs.com/blog/nvidia-h100-gpu-deep-learning-performance-analysis#:~:text=Compared%20to%20NVIDIA's%20previous%2Dgeneration,multiprocessors%2C%20and%20higher%20clock%20frequency.
https://www.patreon.com/AIExplained Non-Hype, Free Newsletter: https://signaltonoise.beehiiv.com/
Learn more about your ad choices. Visit megaphone.fm/adchoices
A head-spinning week of declarations, timelines, updates, papers and debates. I will cover as much as I can, not least the new all-in one interface, EmotionPrompt, executive orders analysed, Bletchley declaration, a new MIT safety paper, responsible scaling, LLMs for Chip Design, and a guest appearance from Dyson Spheres.
https://www.patreon.com/AIExplained
Gates Interview: https://www.handelsblatt.com/technik/ki/bill-gates-mit-ki-koennen-medikamente-viel-schneller-entwickelt-werden/29450298.html
Dwarkesh Interview Legg: https://www.youtube.com/watch?v=Kc1atfJkiJU
Dwarkesh Interview Christiano: https://www.youtube.com/watch?v=9AAhTLa0dT0
Representation Engineering: https://arxiv.org/pdf/2310.01405.pdf
MLC: https://www.scientificamerican.com/article/new-training-method-helps-ai-generalize-like-people-do/
Anthropic Policy: https://www.anthropic.com/index/uk-ai-safety-summit
Science Automation: https://arxiv.org/ftp/arxiv/papers/2304/2304.05332.pdf
https://arxiv.org/pdf/2304.05376.pdf
Jim Fan Tweet: https://twitter.com/DrJimFan/status/1719733318521086332
White House Executive Order: https://www.whitehouse.gov/briefing-room/presidential-actions/2023/10/30/executive-order-on-the-safe-secure-and-trustworthy-development-and-use-of-artificial-intelligence/
ChatGPT Update: https://twitter.com/NorthstarBrain/status/1718708111706636403?s=20
RSP policies by AGI Labs: https://www.aisafetysummit.gov.uk/policy-updates/#company-policies
YouGov Poll: https://d3nkl3psvxxpe9.cloudfront.net/documents/ControlAI_AI_231019.pdf
https://www.patreon.com/AIExplained Non-Hype, Free Newsletter: https://signaltonoise.beehiiv.com/
Learn more about your ad choices. Visit megaphone.fm/adchoices
The first model set to be opensourced that actually comes close to ChatGPT, and is just 13B (that's small enough for a laptop). The 51 page report from Microsoft was released just 48 hours ago but I have gone through it all, and bring relevant insights from 5 other papers.
By imitating the logic and explanations of GPT 4 (and using GPT 3.5 as an assistant), as well as by training on diverse tasks and an order of magnitude more examples, we have Orca. I will showcase it on a dozen benchmarks and go through in detail how it works and why.
I will also end on comments from Sam Altman and Ilya Sutskever on whether Opensource will catch-up...
Orca Paper: https://arxiv.org/pdf/2306.02707.pdf
False Promise Paper: https://arxiv.org/pdf/2305.15717.pdf
https://sharegpt.com/
FLAN: https://arxiv.org/pdf/2301.13688.pdf
Vicuna: https://lmsys.org/blog/2023-03-30-vicuna/
No Moat memo: https://www.semianalysis.com/p/google-we-have-no-moat-and-neither
LLM Leaderboard: https://chat.lmsys.org/?leaderboard
AGIEval: https://arxiv.org/pdf/2304.06364.pdf
BIG-Bench Hard: https://arxiv.org/pdf/2210.09261.pdf
Language Models as Tool Makers: https://arxiv.org/pdf/2305.17126.pdf
Altman Interview: https://www.youtube.com/watch?v=VWUhASix9ws
DERA Paper: https://arxiv.org/abs/2303.17071
Let's Verify Step by Step: https://arxiv.org/abs/2305.20050
https://www.patreon.com/AIExplained Non-Hype, Free Newsletter: https://signaltonoise.beehiiv.com/
Learn more about your ad choices. Visit megaphone.fm/adchoices
With headlines of an imminent job apocalypse, code red for ChatGPT and recursive self-improvement, at the same time as Anthropic's CEO yesterday saying we know how to scale to AGI, and Gemini 3 DeepThink out today, it is easy to get lost among the narratives and counter-narratives. So here are both, plus the facts behind them, for you to decide.
https://epoch.ai/ai-explained-datacenters
Epoch AI is the sponsor of today’s video, and my views, and those expressed in this video, do not necessarily reflect Epoch AI’s views in any way.
Chapters:
00:00 - Introduction
00:42 - Job Apocalypse?
01:45 - Scaling to AGI
04:15 - Recursive Self-Improvement Needed, or Not
09:57 - OpenAI Code Red vs Gemini 3 DeepThink vs Claude Opus 4.5
13:27 - DeepSeek Speciale vs Mistral Large v3
16:45 - Claude Soul Document
https://lmcouncil.ai/
AI Insiders ($9!): https://www.patreon.com/AIExplained
Guardian Interview: https://www.theguardian.com/technology/ng-interactive/2025/dec/02/jared-kaplan-artificial-intelligence-train-itself
MIT Study on Jobs/Tasks: https://iceberg.mit.edu/report.pdf
vs https://www.cnbc.com/2025/11/26/mit-study-finds-ai-can-already-replace-11point7percent-of-us-workforce.html
Amodei on Scaling: https://www.youtube.com/watch?v=FEj7wAjwQIk
Claude Soul Document: https://www.lesswrong.com/posts/vpNG99GhbBoLov9og/claude-4-5-opus-soul-document
Capabilities Original Stance: https://www.anthropic.com/news/core-views-on-ai-safety
Ilya Interview: https://www.dwarkesh.com/p/ilya-sutskever-2
Ricursive Intelligence: https://x.com/RicursiveAI/status/1995932204703346946
Economist Worker Usage of GenAI: https://www.economist.com/finance-and-economics/2025/11/26/investors-expect-ai-use-to-soar-thats-not-happening#selection-1409.94-1413.42
Mistral v3 Large: https://docs.mistral.ai/models/mistral-large-3-25-12
Compute Slowdown Paper: https://joel-becker.com/images/publications/forecasting_time_horizon_under_compute_slowdown.pdf
https://x.com/joel_bkr/status/1993023436541903155
METR Chart: https://metr.org/blog/2025-03-19-measuring-ai-ability-to-complete-long-tasks/
https://www.theinformation.com/articles/openais-350-billion-computing-cost-problem?rc=sy0ihq
OpenAI Code Red: https://www.anthropic.com/news/core-views-on-ai-safety
Rocket Company: https://www.independent.co.uk/news/world/americas/sam-altman-rocket-elon-musk-spacex-b2878351.html
DeepSeek Paper: https://arxiv.org/html/2512.02556v1
DeepSeek Crowdstrike CCP: https://www.crowdstrike.com/en-us/blog/crowdstrike-researchers-identify-hidden-vulnerabilities-ai-coded-software/
https://simple-bench.com/
Patreon Post: https://www.patreon.com/c/aiexplained/posts
Robot: https://x.com/jloganolson/status/1985850115379351799
Non-hype Newsletter: https://signaltonoise.beehiiv.com/
Podcast: https://aiexplainedopodcast.buzzsprout.com/
Learn more about your ad choices. Visit megaphone.fm/adchoices
Gemini 3 Pro is out, and records fell like snowflakes in Svalbard.
No long description, chapters or links today, huge technical difficulties, including with audio, so just want to publish asap.
https://app.grayswan.ai/ai-explained
https://lmcouncil.ai
AI Insiders ($9!): https://www.patreon.com/AIExplained
Non-hype Newsletter: https://signaltonoise.beehiiv.com/
Podcast: https://aiexplainedopodcast.buzzsprout.com/
Learn more about your ad choices. Visit megaphone.fm/adchoices
The two models that you will hear discussed for at least the next two months - Claude Opus 4.6 and GPT 5.3 Codex - just got released within 26 mins or each other. The full breakdown of around 250 pages of reports, with just the most interest moments, from the battle of which is best, Claude personhood, the surprising misbehaviour of Opus 4.6, and much more
https://assemblyai.com/aiexplained
Check out my fast-growing (!) app, free to use, and code INSIDER15 for Pro: https://lmcouncil.ai
AI Insiders ($9): https://www.patreon.com/AIExplained
Chapters:
00:00 - Introduction
00:54 - Self-improvement?
02:44 - Knowledge Work
05:30 - Overly agentic behaviour
09:12 - Who Shouldn’t Use Claude Opus
11:39 - Step-change?
15:09 - Claude’s ‘Personhood’
Hassabis Roadmap: https://www.patreon.com/posts/hassabis-roadmap-149750869
Release of Opus 4.6: https://www.anthropic.com/news/claude-opus-4-6
212 Page System Card: https://www-cdn.anthropic.com/0dd865075ad3132672ee0ab40b05a53f14cf5288.pdf
Claude Code Tip: https://x.com/bcherny/status/2019475897691124107
GPT Codex 5.3: https://openai.com/index/introducing-gpt-5-3-codex/
System Card: https://openai.com/index/gpt-5-3-codex-system-card/
Browse Comp: https://arxiv.org/pdf/2504.12516v1
Finance Agent: https://www.vals.ai/benchmarks/finance_agent
Terminal Bench 2: https://arxiv.org/pdf/2601.11868
Vending Bench: https://andonlabs.com/blog/opus-4-6-vending-bench
My X post: https://x.com/AIExplainedYT/status/2016851303436095647
Anthropic Apology: https://x.com/ch402/status/2014066134194995256/photo/1
Altman rebuttal: https://x.com/sama/status/2019139174339928189
https://x.com/sama/status/2019140276246442089
4% of GitHub: https://x.com/dylan522p/status/2019490550911766763
Non-hype Newsletter: https://signaltonoise.beehiiv.com/
Podcast: https://aiexplainedopodcast.buzzsprout.com/
Learn more about your ad choices. Visit megaphone.fm/adchoices
o1. Strawberry. Q-Star. We finally get answers to what the next generation of LLM reasoning abilities will bring us. And it’s good. Better than I thought. Simple-bench partial results, plus an analysis of the full OpenAI system card, press release, new capabilities, benchmarks and far, far more. It is a step-change, and no, that’s not just hype. Welcome to the o1 era.
AI Insiders: https://www.patreon.com/AIExplained
ChatGPT Status: https://x.com/ChatGPTapp/status/1775684027774836873
Mini, o1: https://x.com/willdepue/status/1834294935497179633/photo/2
System Crad: https://cdn.openai.com/o1-system-card.pdf
o1 Intro: https://openai.com/index/introducing-openai-o1-preview/
o1 Learning to reason: https://openai.com/index/learning-to-reason-with-llms/
Unfaithful ‘Thoughts’: https://www-cdn.anthropic.com/827afa7dd36e4afbb1a49c735bfbb2c69749756e/measuring-faithfulness-in-chain-of-thought-reasoning.pdf
Noam Brown Tweet: https://x.com/polynoamial/status/1834280305076961732/photo/1
Human-level Reasoning? https://x.com/max_a_schwarzer/status/1834278973154754711
Can Do Logic Puzzles Now? https://www.youtube.com/watch?v=3BkQI3nIiB8
Brockman Sabbatical Tweet: https://x.com/gdb/status/1834295775674990676
Still Falls Victim to 9.11: https://x.com/SuvanshSanjeev/status/1834290199276179858
Previous Apollo Research on ‘Deception’: https://arxiv.org/pdf/2311.07590
Simple Bench: https://simple-bench.com/index.html
AlphaCode 2: https://storage.googleapis.com/deepmind-media/AlphaCode2/AlphaCode2_Tech_Report.pdf
Chapters:
00:00-26:55: o1 ChatGPT - OpenAI
My New Coursera Course! The 8 Most Controversial Terms in AI: https://imp.i384100.net/m57g3M
Non-hype Newsletter: https://signaltonoise.beehiiv.com/
Many people expense AI Insiders for work. Feel free to use this template: https://docs.google.com/document/d/1l_0RmVE4SYiAkY63LVvkcnyMwj4v7ZhGOKq77JYgrU8/edit#heading=h.nuqqxloqqhim
AI Insiders: https://www.patreon.com/AIExplained
Learn more about your ad choices. Visit megaphone.fm/adchoices
AlphaEvolve is not the first system to exhibit self-improvement, but it may be the most impressive yet. AI is literally improving the hardware, architectures, data and training methods of AI itself. A deep dive into the paper, drawing on two previous interviews and 5 other papers. Plus a snippet on OpenAI’s new Codex system.
Gray Swan: http://app.grayswan.ai/ai-explained
AI Insiders ($9!): https://www.patreon.com/AIExplained
Chapters:
00:00 - Introduction
00:27 - AlphaEvolve
05:23 - Limitation
06:10 - Achievements
08:21 - Future Improvements
13:30 - Quirks
16:34 - Final Thoughts
AlphaEvolve release: https://deepmind.google/discover/blog/alphaevolve-a-gemini-powered-coding-agent-for-designing-advanced-algorithms/
Paper: https://storage.googleapis.com/deepmind-media/DeepMind.com/Blog/alphaevolve-a-gemini-powered-coding-agent-for-designing-advanced-algorithms/AlphaEvolve.pdf
Terence Tao Quote: https://mathstodon.xyz/@tao/114508029896631083
Nature Article: https://www.nature.com/articles/s41586-022-05172-4
MIT Article: https://www.technologyreview.com/2025/05/14/1116438/google-deepminds-new-ai-uses-large-language-models-to-crack-real-world-problems/
AI Co-Scientist: https://arxiv.org/pdf/2502.18864
OpenAI Codex: https://openai.com/index/introducing-codex/
70% of Pull Requests: https://x.com/slow_developer/status/1920920456393028027
Amodei Essay: https://www.darioamodei.com/essay/machines-of-loving-grace
OpenAI Jason Wei Tweet: https://x.com/_jasonwei/status/1923091260354531612
PromptBreeder: https://arxiv.org/pdf/2309.16797
DrEureka: https://arxiv.org/pdf/2406.01967
FT DeepMind: https://www.ft.com/content/4e497a91-670a-4f69-be4a-18e247daba3e
Non-hype Newsletter: https://signaltonoise.beehiiv.com/
Podcast: https://aiexplainedopodcast.buzzsprout.com/
Learn more about your ad choices. Visit megaphone.fm/adchoices
It’s probably not possible to satisfactorily condense a 12 month’s worth of weird progress in AI, as well as predictions for the year to come, into one video. But I’m gonna try anyway because it has been a very strange time.
http://matsprogram.org/s26-aie
My new app! https://lmcouncil.ai
Patreon Interview: https://www.patreon.com/posts/robot-in-your-27-146376094
Chapters:
00:00 - Introduction
00:34 - Reasoning Models … and limits
02:54 - A playable world
03:36 - Realism
03:50 - AI Slop gone mainstream
05:03 - DolphinGemma
05:39 - Public Mood
07:34 - AI Enlisted
08:30 - GPT-5
11:05 - Open Weight not out
13:00 - METR Breakout
17:30 - VASA-1
18:28 - Lateral Productivity
20:15 - 1 or 1000 benchmarks needed?
24:54 - Continual Learning + Altman on Superintelligence
28:08 - Automated Information Discovery ft AlphaEvolve
Hassabis on Generality: https://x.com/demishassabis/status/2003097405026193809
https://www.youtube.com/watch?v=PqVbypvxDto
Gemini 3: https://storage.googleapis.com/gweb-uniblog-publish-prod/original_images/gemini_3_table_final_HLE_Tools_on.gif
Reasoning Trade-offs: https://arxiv.org/pdf/2504.13837
DolphinGemma: https://blog.google/technology/ai/dolphingemma/?s=09
Genie 3: https://deepmind.google/blog/genie-3-a-new-frontier-for-world-models/
METR Time Horizon: https://arxiv.org/pdf/2503.14499
https://metr.org/blog/2025-03-19-measuring-ai-ability-to-complete-long-tasks/
Flaws: https://x.com/ShashwatGoel7/status/2002369517499105443
https://shash42.substack.com/p/how-to-game-the-metr-plot
https://x.com/METR_Evals/status/2002203627377574113
GPT-5 - Altman phd in everything: https://edition.cnn.com/2025/08/14/business/chatgpt-rollout-problems
https://simple-bench.com/
AI Slop: https://www.youtube.com/watch?v=I_3vxoJDD9k
https://www.theguardian.com/technology/2025/dec/16/boost-for-artists-in-ai-copyright-battle-as-only-3-per-cent-back-uk-active-opt-out-plan
Survey: https://x.com/SearchlightInst/status/2001057144842387920/photo/1
Nvidia Nemotron: https://x.com/percyliang/status/2000608134205985169
OpenAI Compute Flywheel: https://x.com/OpenAI/status/2001363007209914399/photo/1
Altman Interview: https://www.youtube.com/watch?v=2P27Ef-LLuQ
AI in Govt: https://x.com/jdcmedlock/status/1939814516503847259
Benchmark Gaming: https://techcrunch.com/2025/04/07/meta-exec-denies-the-company-artificially-boosted-llama-4s-benchmark-scores/
AlphaEvolve: https://deepmind.google/blog/alphaevolve-a-gemini-powered-coding-agent-for-designing-advanced-algorithms/
https://storage.googleapis.com/deepmind-media/DeepMind.com/Blog/alphaevolve-a-gemini-powered-coding-agent-for-designing-advanced-algorithms/AlphaEvolve.pdf?utm_source=deepmind.google&utm_medium=referral&utm_campaign=gdm&utm_content=
Continual Learning: https://abehrouz.github.io/files/NL.pdf
Job Risk: https://archive.ph/20250708204527/https://www.axios.com/2025/05/28/ai-jobs-white-collar-unemployment-anthropic
GPT4o: https://x.com/AISafetyMemes/status/1916889492172013989
Vasa-1: https://www.microsoft.com/en-us/research/project/vasa-1/
Three Views: https://www.lesswrong.com/posts/K2D45BNxnZjdpSX2j/ai-timelines
Turing Test: https://x.com/tunguz/status/1907185471211422147
Karpathy Year in Review: https://karpathy.bearblog.dev/year-in-review-2025/
LLM Brainrot: https://arxiv.org/pdf/2510.13928
Lateral Productivity: https://www.aisi.gov.uk/frontier-ai-trends-report
Emotional Quotient: https://arxiv.org/pdf/2511.08394
Non-hype Newsletter: https://signaltonoise.beehiiv.com/
Podcast: https://aiexplainedopodcast.buzzsprout.com/
AI Insiders ($9!): https://www.patreon.com/AIExplained
Reasoning Models would boost results but not change paradigm
Genie 3 makes the world playable (literally, in the case of yesterday’s news)
DolphinDecoding
Veo 3.1 / Sora 2 / Nano Banana Pro / Elevenlabs Voice/Music
But AI slop everywhere
Public attitude very mixed (Hassabis)
AI gets e
Learn more about your ad choices. Visit megaphone.fm/adchoices
Tips on how best to use the stunning new Midjourney v6, a look at the new claims of 'imminent' healthspan extension and a glimpse at the potential of Gemini 2, given the compute power of Google - plus Suno AI, RunwayML and more. Welcome to the Christmas Edition of AI Explained!
https://www.patreon.com/AIExplained
Midjourney v6: https://mid-journey.ai/midjourney-v6-release/
Magnific: https://magnific.ai/editor/
RunwayML: https://runwayml.com/
Suno AI: https://app.suno.ai/create/
Altman Blog: https://blog.samaltman.com/what-i-wish-someone-had-told-me
Bloomberg Retro: https://www.bloomberg.com/news/features/2023-12-19/longevity-startup-retro-biosciences-is-sam-altman-s-shot-at-life-extension
Cell Reprogramming: https://www.ncbi.nlm.nih.gov/pmc/articles/PMC9845736/
Big Tech Longevity Funding: https://www.businessinsider.com/richest-wealthiest-entrepreneurs-ceo-billionaires-tech-searching-hacking-longevity?r=US&IR=T#jeff-bezos-has-also-gotten-into-longevity-investment-game-in-recent-years-9
Altman Metformin: https://www.businessinsider.com/openai-ceo-sam-altman-takes-diabetes-drug-metformin-slow-aging-2023-3?r=US&IR=T
And Empty Bank Account: https://www.technologyreview.com/2023/03/08/1069523/sam-altman-investment-180-million-retro-biosciences-longevity-death/
20 Years – Altos Labs: https://english.elpais.com/science-tech/2022-03-08/juan-carlos-izpisua-within-two-decades-we-will-be-able-to-prevent-aging.html
Jerry Tworek OpenAI: https://twitter.com/MillionInt/status/1737578181509648468
Gemini 2: https://www.theinformation.com/articles/how-google-got-back-on-its-feet-in-ai-race?rc=sy0ihq
SemiAnalysis: https://www.semianalysis.com/p/google-gemini-eats-the-world-gemini Non-Hype, Free Newsletter: https://signaltonoise.beehiiv.com/
Learn more about your ad choices. Visit megaphone.fm/adchoices