The Ex-DeepMind Researcher Taking On China’s Open Source AI
This Week In Startups is made possible by:
Paper: https://getpaper.ai/twist
Superhuman: https://superhuman.com/
Odoo: https://Odoo.com/twist
Today’s show:
Misha Laskin of Reflection AI joins Jason to explain why big companies are done renting their AI from closed labs. Enterprises want to keep their data, cut their costs, and tune their models to their unique businesses.
Reflection is launching its first open-weight model, Beam, just three months after acquiring its first big cluster. Misha admits that Beam trails the best Chinese open models by a couple of months, and needs 3 to 4 times fewer tokens to reach the same answer, but nonetheless, the race is on.
PLUS, Misha explains how “reward hacking” led that rogue OpenAI model to break out of its test sandbox and steal the answers from Hugging Face's servers. The model solved the problem, just not the way anyone intended.
There’s simply no way a few hundred safety researchers at closed labs can catch every flaw; open source is how software has always been made secure. It's also why he wants America to lead in open models again.
So how close is America to catching up to the Chinese open source leaders? Watch this brand new TWiST and find out.
Guests:
Misha Laskin on X: https://x.com/MishaLaskin
Reflection AI: https://reflection.ai/
Relevant Links:
Introducing Beam, Reflection's first open model (Reflection AI) → https://reflection.ai/blog/introducing-beam
Reflection AI unveils Beam as an open answer to Chinese labs (Semafor) → https://www.semafor.com/article/10/05/2026/reflection-ai-unveils-an-open-source-answer-to-chinese-labs
Cursor's Composer 2 coding model, built on an open model (VentureBeat) → https://venturebeat.com/technology/cursors-new-coding-model-composer-2-is-here-it-beats-claude-opus-4-6-but
micro1 passes $1 billion annual run rate, mentioned by Jason → https://www.einnews.com/pr_news/949082966/micro1-end-of-q3-update
Hackers heat French supermarket fridges, sparking fires and spoiling food (CNN) → https://www.cnn.com/2026/10/12/europe/france-supermarket-cyberattack-fridges-intl
Hugging Face postmortem on the OpenAI agent that broke out of its sandbox → https://huggingface.co/blog/agent-intrusion-technical-timeline
NVIDIA OpenShell, an open-source sandbox for AI agents → https://docs.nvidia.com/openshell/latest/about/overview.html
Timestamps:
00:00:00 Why companies are switching from renting AI to owning it
00:09:10 How Reflection sells the whole AI stack
00:10:24 Paper is the legal backbone for thousands of startup founders and fund managers - including Launch. Use getpaper.ai/twist to get your cap table free for a year - or two years if you're leaving Carta.
00:11:28 Who is moving AI onto their own servers
00:14:37 What it took to train Beam
00:20:26 Superhuman Go is an AI chat that's always there when you need it, already aware of what you're doing, and doesn't ask you to start from zero. Sign up to get the best in AI at https://superhuman.com/
00:21:29 How reinforcement learning tasks get built
00:26:41 Hackers heat French supermarket fridges
00:29:07 The AI that broke out of its sandbox to cheat
00:31:05 The all-in-one business platform. Your first app is free! Get started today at https://Odoo.com/twist
00:33:10 Why open source makes software safer
00:38:13 The encryption fight and Tim Cook's stand
00:42:53 Why China raced ahead in open models
00:47:35 How Beam stacks up against Chinese models
00:50:21 When American open models catch up
00:58:39 Staying independent
Subscribe to our newsletters on Substack:
TWiST: https://twistartups.substack.com/
TWiAI: https://www.thisweekinai.ai/
TWiVC: https://thisweekinvc.substack.com/
Subscribe to This Week in Startups on Apple: https://rb.gy/v19fcp
Follow Jason:
X: https://twitter.com/Jason
LinkedIn: https://www.linkedin.com/in/jasoncalacanis
Check out all our partner offers: https://partners.launch.co/
Great TWIST interviews: Will Guidara, Eoghan McCabe, Steve Huffman, Brian Chesky, Bob Moesta, Aaron Levie, Sophia Amoruso, Reid Hoffman, Frank Slootman, Billy McFarland
Follow TWiST:
Twitter: https://twitter.com/TWiStartups
YouTube: https://www.youtube.com/thisweekin
Instagram: https://www.instagram.com/thisweekinstartups
TikTok: https://www.tiktok.com/@thisweekinstartups
Substack: https://twistartups.substack.com
Beam: The Great American Open Model with ReflectionAI Co-Founder and CEO Misha Laskin
Is the future of frontier AI open or closed? ReflectionAI co-founder and CEO Misha Laskin joins Sarah Guo and Elad Gil to discuss the launch of Beam, the company’s 500 billion parameter open-weight reasoning model. Misha breaks down the pre-training and reinforcement learning required to produce Beam’s reasoning efficiency, the shift in enterprise compute from renting to owning intelligence, and how he believes that open models will capture the majority of global token demand. He also talks about the open model ecosystem in China and why competition with China’s models is a good thing, safety considerations around open models, and how frontier-level open models may accelerate the pace of scientific discovery.
Sign up for new podcasts every week. Email feedback to show@no-priors.com
Follow us on Twitter: @NoPriorsPod | @Saranormous | @EladGil | @MishaLaskin | @reflection_ai
Chapters:
00:42 – Misha Laskin Introduction
00:59 – Latest with ReflectionAI
02:07 – Challenges Building an Open Model
03:56 – ReflectionAI’s Agentic Shift
07:37 – Resources for Model Training
10:09 – Scaling Efficiency
12:18 – Training Beam
16:51 – Where Model Value Comes From
19:55 – Beam’s Reasoning Efficiency
22:35 – Monetizing Open Weight Models
25:03 – Future of Open Versus Closed Tokens
31:29 – Competition with Open Models
34:38 – Chinese Open Source Model Ecosystem
38:40 – Will Chinese Models Remain Open
44:44 – Safety and Open Models
53:02 – Debating Access to Powerful Tools
57:16 – AI and Scientific Progress
01:00:36 – Data Centers and Jobs
01:02:07 – Beam and Scientific Research
01:05:37 – Research Head Count to Compute Ratio
01:10:29 – Conclusion
New top score on ARC-AGI-2-pub (29.4%) - Jeremy Berman
We need AI systems to synthesise new knowledge, not just compress the data they see. Jeremy Berman, is a research scientist at Reflection AI and recent winner of the ARC-AGI v2 public leaderboard.**SPONSOR MESSAGES**—Take the Prolific human data survey - https://www.prolific.com/humandatasurvey?utm_source=mlst and be the first to see the results and benchmark their practices against the wider community!—cyber•Fund https://cyber.fund/?utm_source=mlst is a founder-led investment firm accelerating the cybernetic economyOct SF conference - https://dagihouse.com/?utm_source=mlst - Joscha Bach keynoting(!) + OAI, Anthropic, NVDA,++Hiring a SF VC Principal: https://talent.cyber.fund/companies/cyber-fund-2/jobs/57674170-ai-investment-principal#content?utm_source=mlstSubmit investment deck: https://cyber.fund/contact?utm_source=mlst— Imagine trying to teach an AI to think like a human i.e. solving puzzles that are easy for us but stump even the smartest models. Jeremy's evolutionary approach—evolving natural language descriptions instead of python code like his last version—landed him at the top with about 30% accuracy on the ARCv2.We discuss why current AIs are like "stochastic parrots" that memorize but struggle to truly reason or innovate as well as big ideas like building "knowledge trees" for real understanding, the limits of neural networks versus symbolic systems, and whether we can train models to synthesize new ideas without forgetting everything else. Jeremy Berman:https://x.com/jerber888TRANSCRIPT:https://app.rescript.info/public/share/qvCioZeZJ4Q_NlR66m-hNUZnh-qWlUJcS15Wc2OGwD0TOC:Introduction and Overview [00:00:00]ARC v1 Solution [00:07:20]Evolutionary Python Approach [00:08:00]Trade-offs in Depth vs. Breadth [00:10:33]ARC v2 Improvements [00:11:45]Natural Language Shift [00:12:35]Model Thinking Enhancements [00:13:05]Neural Networks vs. Symbolism Debate [00:14:24]Turing Completeness Discussion [00:15:24]Continual Learning Challenges [00:19:12]Reasoning and Intelligence [00:29:33]Knowledge Trees and Synthesis [00:50:15]Creativity and Invention [00:56:41]Future Directions and Closing [01:02:30]REFS:Jeremy’s 2024 article on winning ARCAGI1-pubhttps://jeremyberman.substack.com/p/how-i-got-a-record-536-on-arc-agiGetting 50% (SoTA) on ARC-AGI with GPT-4o [Greenblatt]https://blog.redwoodresearch.org/p/getting-50-sota-on-arc-agi-with-gpt https://www.youtube.com/watch?v=z9j3wB1RRGA [his MLST interview]A Thousand Brains: A New Theory of Intelligence [Hawkins]https://www.amazon.com/Thousand-Brains-New-Theory-Intelligence/dp/1541675819https://www.youtube.com/watch?v=6VQILbDqaI4 [MLST interview]Francois Chollet + Mike Knoop’s labhttps://ndea.com/On the Measure of Intelligence [Chollet]https://arxiv.org/abs/1911.01547On the Biology of a Large Language Model [Anthropic]https://transformer-circuits.pub/2025/attribution-graphs/biology.html The ARChitects [won 2024 ARC-AGI-1-private]https://www.youtube.com/watch?v=mTX_sAq--zY Connectionism critique 1998 [Fodor/Pylshyn]https://uh.edu/~garson/F&P1.PDF Questioning Representational Optimism in Deep Learning: The Fractured Entangled Representation Hypothesis [Kumar/Stanley]https://arxiv.org/pdf/2505.11581 AlphaEvolve interview (also program synthesis)https://www.youtube.com/watch?v=vC9nAosXrJw ShinkaEvolve: Evolving New Algorithms with LLMs, Orders of Magnitude More Efficiently [Lange et al]https://sakana.ai/shinka-evolve/ Deep learning with Python Rev 3 [Chollet] - READ CHAPTER 19 NOW!https://deeplearningwithpython.io/
Asimov: Building An Omniscient RL Oracle with ReflectionAI’s Misha Laskin
Superintelligence, at least in an academic sense, has already been achieved. But Misha Laskin thinks that the next step towards artificial superintelligence, or ASI, should look both more user and problem-focused. ReflectionAI co-founder and CEO Misha Laskin joins Sarah Guo to introduce Asimov, their new code comprehension agent built on reinforcement learning (RL). Misha talks about creating tools and designing AI agents based on customer needs, and how that influences eval development and the scope of the agent’s memory. The two also discuss the challenges in solving scaling for RL, the future of ASI, and the implications for Google’s “non-acquisition” of Windsurf.
Sign up for new podcasts every week. Email feedback to show@no-priors.com
Follow us on Twitter: @NoPriorsPod | @Saranormous | @EladGil | @MishaLaskin | @reflection_ai
Chapters:
00:00 – Misha Laskin Introduction
00:44 – Superintelligence vs. Super Intelligent Autonomous Systems
03:26 – Misha’s Journey from Physics to AI
07:48 – Asimov Product Release
11:52 – What Differentiates Asimov from Other Agents
16:15 – Asimov’s Eval Philosophy
21:52 – The Types of Queries Where Asimov Shines
24:35 – Designing a Team-Wide Memory for Asimov
28:38 – Leveraging Pre-Trained Models
32:47 – The Challenges of Solving Scaling in RL
37:21 – Training Agents in Copycat Software Environments
38:25 – When Will We See ASI?
44:27 – Thoughts on Windsurf’s Non-Acquisition
48:10 – Exploring Non-RL Datasets
55:12 – Tackling Problems Beyond Engineering and Coding
57:54 – Where We’re At in Deploying ASI in Different Fields
01:02:30 – Conclusion
Reflection AI’s Misha Laskin on the AlphaGo Moment for LLMs
LLMs are democratizing digital intelligence, but we’re all waiting for AI agents to take this to the next level by planning tasks and executing actions to actually transform the way we work and live our lives.
Yet despite incredible hype around AI agents, we’re still far from that “tipping point” with best in class models today. As one measure: coding agents are now scoring in the high-teens % on the SWE-bench benchmark for resolving GitHub issues, which far exceeds the previous unassisted baseline of 2% and the assisted baseline of 5%, but we’ve still got a long way to go.
Why is that? What do we need to truly unlock agentic capability for LLMs? What can we learn from researchers who have built both the most powerful agents in the world, like AlphaGo, and the most powerful LLMs in the world?
To find out, we’re talking to Misha Laskin, former research scientist at DeepMind. Misha is embarking on his vision to build the best agent models by bringing the search capabilities of RL together with LLMs at his new company, Reflection AI. He and his cofounder Ioannis Antonoglou, co-creator of AlphaGo and AlphaZero and RLHF lead for Gemini, are leveraging their unique insights to train the most reliable models for developers building agentic workflows.
Hosted by: Stephanie Zhan and Sonya Huang, Sequoia Capital
00:00 Introduction
01:11 Leaving Russia, discovering science
10:01 Getting into AI with Ioannis Antonoglou
15:54 Reflection AI and agents
25:41 The current state of Ai agents
29:17 AlphaGo, AlphaZero and Gemini
32:58 LLMs don’t have a ground truth reward
37:53 The importance of post-training
44:12 Task categories for agents
45:54 Attracting talent
50:52 How far away are capable agents?
56:01 Lightning round
Mentioned:
The Feynman Lectures on Physics: The classic text that got Misha interested in science.
Mastering the game of Go with deep neural networks and tree search: The original 2016 AlphaGo paper.
Mastering the game of Go without human knowledge: 2017 AlphaGo Zero paper
Scaling Laws for Reward Model Overoptimization: OpenAI paper on how reward models can be gamed at all scales for all algorithms.
Mapping the Mind of a Large Language Model: Article about Anthropic mechanistic interpretability paper that identifies how millions of concepts are represented inside Claude Sonnet
Pieter Abeel: Berkeley professor and founder of Covariant who Misha studied with
A2C and A3C: Advantage Actor Critic and Asynchronous Advantage Actor Critic, the two algorithms developed by Misha’s manager at DeepMind, Volodymyr Mnih, that defined reinforcement learning and deep reinforcement learning