martin goodson

The London Machine Learning Meetup is the largest machine learning community in Europe. Previous speakers include Juergen Schmidhuber, David Silver, Yoshua Bengio and Andrej Karpathy. Join our next meetup: meetup.com/London-Machine… Or watch videos: youtube.com/c/LondonMachin…

@ipeirotis It was not clear from the post where the oral exam took place? At the students home on a video call or in -person?

posted: "Fighting Fire with Fire: Scalable Oral Exams with an ElevenLabs Voice AI Agent" How we changed our final exam from the traditional written format to an AI-powered oral exam, using ElevenLabs voice agent, and how we used an "LLM council" approach to grade and give feedback to the students. For details: behind-the-enemy-lines.com/2025/12/fighti…

@jamieamartin1 The studies you cite are about the negative effects of childcare for children of age 0-2. Free childcare in the uk is for children of 3 and 4. Please cite appropriate evidence for your claim.

@Support We need to regain access to our companies account @EvolutionAI. We have tried everything but the email codes never reach our email inbox. Please help.

Did the White House use ChatGPT Vibe Code a retarded formula for our tariffs and then were too dumb to cover it up? Yes. Here's the wildest trade story you missed today 🧵👇

@pfau They've been doing stupid stuff with machine learning for a long time on.ft.com/4ideBUA

I don't even work on the kind of AI that they are talking about here, but shit like this makes me seriously consider quitting the whole field for good. It's not the fun, curiosity driven, open field I joined 15 years ago. It's all much darker now.

Erm, can't you just look at the rate of innovation?

@Scobleizer Unfortunately the accuracy is about 80%. I wouldn't be surprised if a single feature (eg feeding problems) was responsible for all of the predictive power of this "AI" system.

We saw many signs in our own autistic son, but it was when he ran away from us without looking back that we really knew we couldn't ignore it any longer. He was just about two. If humans can see it, AI can see it sooner. At least that's my philosophy and it is quickly becoming true.

Our CEO, Dr @martingoodson, will participate in @ArmstrongWolfe & @LSEGplc's COO Debating Society event next week. The motion of the debate is 'AI presents a greater opportunity than threat to Financial Services'. If you have any insights, please leave them below.

If OpenAI has no moat, why is google struggling to compete with them?

Task contamination is worse than we imagined? Models perform better on datasets that were released before the models + Models can generate examples from tasks (NYT vibes) Changmao Li @jmflanig arxiv.org/abs/2312.16337

No moat memo continues to age well. Open source will inevitably take over AI. There is no AI without open source anyway. You have no OpenAI without Linux, containers, Pytorch, Transformers, etc. Only thing that can kill it is regulatory assassination. semianalysis.com/p/google-we-ha…

Sharing my recent slides on how to train and deploy open-source LLMs Goes over * SFT, DPO with tools like @huggingface TRL/PEFT * renting GPUs on @runpod and others * setting up serverless vs dedicated APIs with TGI/vLLM/@togethercompute etc. docs.google.com/presentation/d…

Ok, by popular demand: a starter set of papers you can read on the topic. "Comparing Humans, GPT-4, and GPT-4V On Abstraction and Reasoning Tasks": arxiv.org/abs/2311.09247 "Embers of Autoregression: Understanding Large Language Models Through the Problem They are Trained to Solve": arxiv.org/abs/2309.13638 "Faith and Fate: Limits of Transformers on Compositionality": arxiv.org/abs/2305.18654 "The Reversal Curse: LLMs trained on "A is B" fail to learn 'B is A'": arxiv.org/abs/2309.12288 "On the measure of intelligence": arxiv.org/abs/1911.01547 not about LLMs, but provides context and grounding on what it means to be intelligent and the nature of generalization. It also introduces an intelligence benchmark (ARC) that remains completely out of reach for LLMs. Ironically the best-performing LLM-based systems on ARC are those that have been trained on tons of generated tasks, hoping to hit some overlap between test set tasks and your generated tasks -- LLMs have zero ability to tackle an actually new task. In general there's a new paper documenting the lack of broad generalization capabilities of LLMs every few days.

📸 Look at this post on Facebook by Tomas Mikolov facebook.com/share/p/X2y8F6…

@erikbryn The technique in this paper is more suited to algorithm discovery rather than scientific discovery. It relies on having a method for verifying a proposed solution. The LLM doesn't know the answer, but it can generate good proposals. 1/

magnet:?xt=urn:btih:5546272da9065eddeb6fcd7ffddeef5b75be79a7&dn=mixtral-8x7b-32kseqlen&tr=udp%3A%2F%2Fopentracker.i2p.rocks%3A6969%2Fannounce&tr=http%3A%2F%https://t.co/g0m9cEUz0T%3A80%2Fannounce RELEASE a6bbd9affe0c2725c1b7410d66833e24

I've discovered a Brexit opportunity. The firm founded by Jacob Rees-Mogg is winding down, after St James's Place severed ties - amid regulatory pressure to comply with new FCA consumer duty rules. Amazingly, the FCA couldn't have imposed these rules before Brexit.

@DataJunkie Maybe one of the millions of starving people who got to eat some food because of it?

Is there anyone out there that actually feels that "Do They Know It's Christmas?" by Band Aid is not a tone deaf, offensive and patronizing trash heap of a song? youtube.com/watch?v=j3fSkn…

@sarahookr Maybe this? arxiv.org/abs/2311.09766

I seem to remember a paper that just came out a few weeks ago showing gpt preferred outputs similar to its own distribution. Does anyone know which paper I'm talking about?

This is doing numbers at the moment but the data does not show that TikTok *drives* antisemitism. This is a clear case of selection bias: people who use TikTok are from a different demographic to those who don't. This is very disappointing. @antgoldbloom should know better.