Machine Learning Street Talk
@MLStreetTalk • 39,324 subscribers
MLST is by Dr. Tim Scarfe @ecsquendor w/ cameos from @DoctorDuggar https://t.co/5YCv2SdFwN (early access/priv.discord) - Sponsor us!
Videos
A masterclass from Jeremy Howard on why AI coding tools can be a trap -- and what 45 years of programming taught him that most vibe coders will never learn. - AI coding tools exploit gambling psychology - The difference between typing code and software engineering - Enterprise coding AND prompt-only vibe coding are "inhumane" i.e. disconnecting humans from understanding-building - AI tools remove the "desirable difficulty" you need to build deep mental models. Out on MLST now!
Machine Learning Street Talk170,249 просмотров • 6 месяцев назад
An apparently AI-generated formal proof, in Lean, purporting to be a disproof to the Collatz conjecture, was actually exploiting a bug in the Lean kernel! EXCLUSIVE from Lean Creator Leo de Moura: > "This is going to keep happening. AIs are really good at exploiting soundness bugs in the kernels." It appeared to pass both Lean's official kernel and Nanoda by hitting two separate bugs. Both have now been patched - Leonardo de Moura.
Machine Learning Street Talk34,076 просмотров • 1 месяц назад
One of the most exciting new AI companies is Sakana AI in our opinion. Their CTO and their Co-founder Llion Jones Llion Jones is now saying it's time to move beyond transformers even though he was one of the 8 original inventors at Google. They are now investigating the next significant step forwards, and have a NeurIPS 2025 spotlight paper called "continuous thought machines" (CTM) which may well be just that (first author Luke Darlow). They just landed their Series B, with the legendary hardmaru as their CEO. They are actively investigating neuroevolution approaches which I strongly believe will be a big part of the future of AI -- and they publish their research in the open. Keep an eye on Sakana! 👌 Interview with Llion and Luke dropping today on MLST.
Machine Learning Street Talk187,304 просмотров • 9 месяцев назад
John Jumper, on why AlphaFold2 worked: > "It's not one or two home runs. It's 18 doubles. Midsize wins stacked together." His least popular take on AlphaFold2 is the over-crediting of SE(3)-equivariance / geometric deep learning. He ablated it and it only lifted performance about 2/30 points. FAPE (the frame-aligned point error) was actually most responsible for its improvement.
Machine Learning Street Talk52,796 просмотров • 2 месяцев назад
EMERGENCY PODCAST: The recent paper on the reasoning heist went super viral. It was also 100+ pages long and you didn't have time to read it. The authors Ilia Shumailov🦔 and Alexander Panfilov unpack the paper and discuss model distillation.
Machine Learning Street Talk16,047 просмотров • 20 дней назад
Michael I. Jordan on the new MLST. Four things: > AGI is a PR term. It confuses young people. > Discourse is bipolar, either alarmist or exuberant, this is in his words "so demoralizing" for 20- and 25-year-old researchers. > ML's methods came from statistics and operations research, NOT the AI tradition. > Data markets are Stackelberg games, not optimisation problems. A lot of ML researchers have never computed an equilibrium. Michael I. Jordan is a no-nonsense original gangster of the field and was described by Science magazine, back in 2016 as the most influential living computer scientist.
Machine Learning Street Talk60,343 просмотров • 3 месяцев назад
Meet the man tasked with building UK sovereign AI by the end of 2026. Cosine has a government-backed mandate to build Britain's first sovereign LLM, training on Isambard. He said it "it boils his blood" that folks in the UK are now second-class AI citizens and that "we have no choice but to make it happen." This is Alistair from Cosine This has become a desperate situation and a matter of the utmost urgency.
Machine Learning Street Talk38,013 просмотров • 2 месяцев назад
ARC-AGI-3 is built different, it has dumbfounded almost all regular attempts so far because it's so much harder than anything that came before. It has no rules, it's agentic and has no explicit goals, they need to be discovered. Tufalabs won the first milestone of ARC Prize > There is no language built into the benchmark, but these guys "put the language back in", because in their view - it's the best way to climb up the notional "abstraction mountain" and effectively use many of the abstractions which have evolved over millions of years of language evolution. > They built a novel harness "The Duck" around a 27B open weights model (Qwen 3.6) to solve extremely challenging and novel reasoning problems that require abstraction. > This is the launch video of their winning agentic harness, "The Duck". We have also released an exclusive interview with them on MLST, just dropped. > The million dollar question is: what will François Chollet think about how they've done it, and is this a step towards AGI?
Machine Learning Street Talk35,121 просмотров • 2 месяцев назад
Professor Yi Ma (Yi Ma) discusses his new open-source book "Learning Deep Representations of Data Distributions" The thesis: Intelligence -- natural or artificial -- can be derived from just TWO principles: Parsimony and Self-Consistency.
Machine Learning Street Talk58,270 просмотров • 8 месяцев назад
AI is so smart, why are its internals 'spaghetti'? We spoke with Kenneth Stanley and Akarsh Kumar (MIT) about their new paper: Questioning Representational Optimism in Deep Learning: The Fractured Entangled Representation Hypothesis. Co-authors: Jeff Clune Joel Lehman
Machine Learning Street Talk84,473 просмотров • 1 год назад
Iman Mirzadeh from Apple Iman Mirzadeh wrote the famous GSM-Symbolic paper a couple of months back which argued that LLMs are learning surface statistics and not genuinely reasoning due to their sensitivity to distractors and out of distribution examples.
Machine Learning Street Talk87,677 просмотров • 1 год назад
The Kaleidoscope Hypothesis by François Chollet
Machine Learning Street Talk41,242 просмотров • 7 месяцев назад
We just released our interview with the father of Generative AI - Jürgen Schmidhuber! The G, P, and T in "ChatGPT" (GPT means "Generative Pre-Trained Transformer") go back to Juergen's work of 1990-91 when he published what's now called "Unnormalised Linear Transformers," "Self-Supervised Pre-Training" for deep learning with long texts, and "Generative Adversarial Networks" for Artificial Curiosity. Remarkably, principles of both Transformers and LSTMs date back to 1991, the only palindromic year of the 20th century! Transformers are easier to parallelise, but LSTMs can solve problems which are unsolvable by Transformers. In this first part of our two part show, we discuss the history and the future of the field, with a focus on abstract planning, reasoning, and "learning to think." It's just dropped on MLST!
Machine Learning Street Talk100,064 просмотров • 2 лет назад
I finally got to meet François Chollet in person recently to interview him about ARC Prize, intelligence vs memorization, human cognitive development, learning abstractions, limits of pattern recognition and consciousness development. These are the best bits. Full show released tomorrow
Machine Learning Street Talk79,058 просмотров • 1 год назад
Today Google DeepMind released AlphaEvolve: a Gemini coding agent for algorithm discovery. It beat the famous Strassen algorithm for matrix multiplication set 56 years ago. Google has been killing it recently. We had early access to the paper and interviewed the researchers.
Machine Learning Street Talk58,517 просмотров • 1 год назад
> Natural data is "generated" from a constrained hierarchical / compositional function. > Deep networks learns that hidden structure from polynomially few examples and creatively generate exponentially many valid new ones. > The depth of the network is what's important to overcome the curse of dimensionality, and potentially invalidate Chomsky's poverty of stimulus argument. > Prof Matthieu wyart, a physicist (Johns Hopkins / EPFL) was the senior author of the Random Hierarchy Model.
Machine Learning Street Talk13,577 просмотров • 2 месяцев назад