Machine Learning Street Talk (MLST)

• AI Agents Can Code 10,000 Lines of Hacking Tools In Seconds - Dr. Ilia Shumailov (ex-GDM)

  Dr. Ilia Shumailov - Former DeepMind AI Security Researcher, now building security tools for AI agentsEver wondered what happens when AI agents start talking to each other—or worse, when they start breaking things? Ilia Shumailov spent years at DeepMind thinking about exactly these problems, and he's here to explain why securing AI is way harder than you think.**SPONSOR MESSAGES**—Check out notebooklm for your research project, it's really powerfulhttps://notebooklm.google.com/—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— We're racing toward a world where AI agents will handle our emails, manage our finances, and interact with sensitive data 24/7. But there is a problem. These agents are nothing like human employees. They never sleep, they can touch every endpoint in your system simultaneously, and they can generate sophisticated hacking tools in seconds. Traditional security measures designed for humans simply won't work.Dr. Ilia Shumailovhttps://x.com/iliaishackedhttps://iliaishacked.github.io/https://sequrity.ai/TRANSCRIPT:https://app.rescript.info/public/share/dVGsk8dz9_V0J7xMlwguByBq1HXRD6i4uC5z5r7EVGMTOC:00:00:00 - Introduction & Trusted Third Parties via ML00:03:45 - Background & Career Journey00:06:42 - Safety vs Security Distinction00:09:45 - Prompt Injection & Model Capability00:13:00 - Agents as Worst-Case Adversaries00:15:45 - Personal AI & CAML System Defense00:19:30 - Agents vs Humans: Threat Modeling00:22:30 - Calculator Analogy & Agent Behavior00:25:00 - IMO Math Solutions & Agent Thinking00:28:15 - Diffusion of Responsibility & Insider Threats00:31:00 - Open Source Security Concerns00:34:45 - Supply Chain Attacks & Trust Issues00:39:45 - Architectural Backdoors00:44:00 - Academic Incentives & Defense Work00:48:30 - Semantic Censorship & Halting Problem00:52:00 - Model Collapse: Theory & Criticism00:59:30 - Career Advice & Ross Anderson TributeREFS:Lessons from Defending Gemini Against Indirect Prompt Injectionshttps://arxiv.org/abs/2505.14534Defeating Prompt Injections by Design. Debenedetti, E., Shumailov, I., Fan, T., Hayes, J., Carlini, N., Fabian, D., Kern, C., Shi, C., Terzis, A., & Tramèr, F. https://arxiv.org/pdf/2503.18813Agentic Misalignment: How LLMs could be insider threatshttps://www.anthropic.com/research/agentic-misalignmentSTOP ANTHROPOMORPHIZING INTERMEDIATE TOKENS AS REASONING/THINKING TRACES!Subbarao Kambhampati et alhttps://arxiv.org/pdf/2504.09762Meiklejohn, S., Blauzvern, H., Maruseac, M., Schrock, S., Simon, L., & Shumailov, I. (2025). Machine learning models have a supply chain problem. https://arxiv.org/abs/2505.22778 Gao, Y., Shumailov, I., & Fawaz, K. (2025). Supply-chain attacks in machine learning frameworks. https://openreview.net/pdf?id=EH5PZW6aCrApache Log4j Vulnerability Guidancehttps://www.cisa.gov/news-events/news/apache-log4j-vulnerability-guidance Bober-Irizar, M., Shumailov, I., Zhao, Y., Mullins, R., & Papernot, N. (2022). Architectural backdoors in neural networks. https://arxiv.org/pdf/2206.07840Position: Fundamental Limitations of LLM Censorship Necessitate New ApproachesDavid Glukhov, Ilia Shumailov, ...https://proceedings.mlr.press/v235/glukhov24a.html AlphaEvolve MLST interview [Matej Balog, Alexander Novikov]https://www.youtube.com/watch?v=vC9nAosXrJw

  --------  

  1:01:07

  --------

  1:01:07
• 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/

  --------  

  1:08:27

  --------

  1:08:27
• Deep Learning is Not So Mysterious or Different - Prof. Andrew Gordon Wilson (NYU)

  Professor Andrew Wilson from NYU explains why many common-sense ideas in artificial intelligence might be wrong. For decades, the rule of thumb in machine learning has been to fear complexity. The thinking goes: if your model has too many parameters (is "too complex") for the amount of data you have, it will "overfit" by essentially memorizing the data instead of learning the underlying patterns. This leads to poor performance on new, unseen data. This is known as the classic "bias-variance trade-off" i.e. a balancing act between a model that's too simple and one that's too complex.**SPONSOR MESSAGES**—Tufa AI Labs is an AI research lab based in Zurich. **They are hiring ML research engineers!** This is a once in a lifetime opportunity to work with one of the best labs in EuropeContact Benjamin Crouzier - https://tufalabs.ai/ —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— Description Continued:Professor Wilson challenges this fundamental belief (fearing complexity). He makes a few surprising points:**Bigger Can Be Better**: massive models don't just get more flexible; they also develop a stronger "simplicity bias". So, if your model is overfitting, the solution might paradoxically be to make it even bigger.**The "Bias-Variance Trade-off" is a Misnomer**: Wilson claims you don't actually have to trade one for the other. You can have a model that is incredibly expressive and flexible while also being strongly biased toward simple solutions. He points to the "double descent" phenomenon, where performance first gets worse as models get more complex, but then surprisingly starts getting better again.**Honest Beliefs and Bayesian Thinking**: His core philosophy is that we should build models that honestly represent our beliefs about the world. We believe the world is complex, so our models should be expressive. But we also believe in Occam's razor—that the simplest explanation is often the best. He champions Bayesian methods, which naturally balance these two ideas through a process called marginalization, which he describes as an automatic Occam's razor.TOC:[00:00:00] Introduction and Thesis[00:04:19] Challenging Conventional Wisdom[00:11:17] The Philosophy of a Scientist-Engineer[00:16:47] Expressiveness, Overfitting, and Bias[00:28:15] Understanding, Compression, and Kolmogorov Complexity[01:05:06] The Surprising Power of Generalization[01:13:21] The Elegance of Bayesian Inference[01:33:02] The Geometry of Learning[01:46:28] Practical Advice and The Future of AIProf. Andrew Gordon Wilson:https://x.com/andrewgwilshttps://cims.nyu.edu/~andrewgw/https://scholar.google.com/citations?user=twWX2LIAAAAJ&hl=en https://www.youtube.com/watch?v=Aja0kZeWRy4 https://www.youtube.com/watch?v=HEp4TOrkwV4 TRANSCRIPT:https://app.rescript.info/public/share/H4Io1Y7Rr54MM05FuZgAv4yphoukCfkqokyzSYJwCK8Hosts:Dr. Tim Scarfe / Dr. Keith Duggar (MIT Ph.D)REFS:Deep Learning is Not So Mysterious or Different [Andrew Gordon Wilson]https://arxiv.org/abs/2503.02113Bayesian Deep Learning and a Probabilistic Perspective of Generalization [Andrew Gordon Wilson, Pavel Izmailov]https://arxiv.org/abs/2002.08791Compute-Optimal LLMs Provably Generalize Better With Scale [Marc Finzi, Sanyam Kapoor, Diego Granziol, Anming Gu, Christopher De Sa, J. Zico Kolter, Andrew Gordon Wilson]https://arxiv.org/abs/2504.15208

  --------  

  2:03:48

  --------

  2:03:48
• Karl Friston - Why Intelligence Can't Get Too Large (Goldilocks principle)

  In this episode, hosts Tim and Keith finally realize their long-held dream of sitting down with their hero, the brilliant neuroscientist Professor Karl Friston. The conversation is a fascinating and mind-bending journey into Professor Friston's life's work, the Free Energy Principle, and what it reveals about life, intelligence, and consciousness itself.**SPONSORS**Gemini CLI is an open-source AI agent that brings the power of Gemini directly into your terminal - https://github.com/google-gemini/gemini-cli--- 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***They kick things off by looking back on the 20-year journey of the Free Energy Principle. Professor Friston explains it as a fundamental rule for survival: all living things, from a single cell to a human being, are constantly trying to make sense of the world and reduce unpredictability. It’s this drive to minimize surprise that allows things to exist and maintain their structure.This leads to a bigger question: What does it truly mean to be "intelligent"? The group debates whether intelligence is everywhere, even in a virus or a plant, or if it requires a certain level of complexity. Professor Friston introduces the idea of different "kinds" of things, suggesting that creatures like us, who can model themselves and think about the future, possess a unique and "strange" kind of agency that sets us apart.From intelligence, the discussion naturally flows to the even trickier concept of consciousness. Is it the same as intelligence? Professor Friston argues they are different. He explains that consciousness might emerge from deep, layered self-awareness—not just acting, but understanding that you are the one causing your actions and thinking about your place in the world.They also explore intelligence at different sizes. Is a corporation intelligent? What about the entire planet? Professor Friston suggests there might be a "Goldilocks zone" for intelligence. It doesn't seem to exist at the super-tiny atomic level or at the massive scale of planets and solar systems, but thrives in the complex middle-ground where we live.Finally, they tackle one of the most pressing topics of our time: Can we build a truly conscious AI? Professor Friston shares his doubts about whether our current computers are capable of a feat like that. He suggests that genuine consciousness might require a different kind of "mortal" computation, where the machine's physical body and its "mind" are inseparable, much like in biological creatures.TRANSCRIPT:https://app.rescript.info/public/share/FZkF8BO7HMt9aFfu2_q69WGT_ZbYZ1VVkC6RtU3eeOITOC:00:00:00: Introduction & Retrospective on the Free Energy Principle00:09:34: Strange Particles, Agency, and Consciousness00:37:45: The Scale of Intelligence: From Viruses to the Biosphere01:01:35: Modelling, Boundaries, and Practical Application01:21:12: Conclusion

  --------  

  1:21:39

  --------

  1:21:39
• The Day AI Solves My Puzzles Is The Day I Worry (Prof. Cristopher Moore)

  We are joined by Cristopher Moore, a professor at the Santa Fe Institute with a diverse background in physics, computer science, and machine learning.The conversation begins with Cristopher, who calls himself a "frog" explaining that he prefers to dive deep into specific, concrete problems rather than taking a high-level "bird's-eye view". They explore why current AI models, like transformers, are so surprisingly effective. Cristopher argues it's because the real world isn't random; it's full of rich structures, patterns, and hierarchies that these models can learn to exploit, even if we don't fully understand how.**SPONSORS**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 economy.Oct 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***Cristopher Moore:https://sites.santafe.edu/~moore/TOC:00:00:00 - Introduction00:02:05 - Meet Christopher Moore: A Frog in the World of Science00:05:14 - The Limits of Transformers and Real-World Data00:11:19 - Intelligence as Creative Problem-Solving00:23:30 - Grounding, Meaning, and Shared Reality00:31:09 - The Nature of Creativity and Aesthetics00:44:31 - Computational Irreducibility and Universality00:53:06 - Turing Completeness, Recursion, and Intelligence01:11:26 - The Universe Through a Computational Lens01:26:45 - Algorithmic Justice and the Need for TransparencyTRANSCRIPT: https://app.rescript.info/public/share/VRe2uQSvKZOm0oIBoDsrNwt46OMCqRnShVnUF3qyoFkFilmed at DISI (Diverse Intelligences Summer Institute)https://disi.org/REFS:The Nature of computation [Chris Moore]https://nature-of-computation.org/ Birds and Frogs [Freeman Dyson]https://www.ams.org/notices/200902/rtx090200212p.pdf Replica Theory [Parisi et al]https://arxiv.org/pdf/1409.2722 Janossy pooling [Fabian Fuchs]https://fabianfuchsml.github.io/equilibriumaggregation/ Cracking the cryptic [YT channel]https://www.youtube.com/c/CrackingTheCrypticSudoko Bench [Sakana]https://sakana.ai/sudoku-bench/Fractured entangled representations “phylogenetic locking in comment” [Kumar/Stanley]https://arxiv.org/pdf/2505.11581 (see our shows on this)The War Against Cliché: [Martin Amis]https://www.amazon.com/War-Against-Cliche-Reviews-1971-2000/dp/0375727167Rule 110 (CA)https://mathworld.wolfram.com/Rule150.htmlUniversality in Elementary Cellular Automata [Matt Cooke]https://wpmedia.wolfram.com/sites/13/2018/02/15-1-1.pdf Small Semi-Weakly Universal Turing Machines [Damien Woods] https://tilde.ini.uzh.ch/users/tneary/public_html/WoodsNeary-FI09.pdf COMPUTING MACHINERY AND INTELLIGENCE [Turing, 1950]https://courses.cs.umbc.edu/471/papers/turing.pdf Comment on Space Time as a causal set [Moore, 88]https://sites.santafe.edu/~moore/comment.pdf Recursion Theory on the Reals and Continuous-time Computation [Moore, 96]

  --------  

  1:34:52

  --------

  1:34:52

Más podcasts de Tecnología

Podcasts a la moda de Tecnología

Acerca de Machine Learning Street Talk (MLST)