When we eventually look back at this early AI era, we'll see that some of the dumbest advice was "don't learn coding" or "don't learn to draw" or "don't learn to write."

Always keep learning.

No matter what happens, keep learning.

Thanks for the kind words.

Small snippet of my holiday reading. What's on your list?

They just need an AI editor too and the ability to download 30 years of experience magically. :)

Seems a bit early to call this, no?

Fine-Tuning Llama 3.1 with SWIFT, Unsloth Alternative for Multi-GPU LLM Training | Shelpuk AI Technology Consulting Learn how to fine-tune the Llama 3.1 model with SWIFT for efficient multi-GPU training. This guide covers everything from setting up a training environment on platforms like RunPod and Google Colab to...

Great write up on my new fav fine tuning, training + inf engine: ms-swift from Alibaba group. Supports every freaking model that matters, a massive amount of full + PEFT finetuning, has tons of backends wrapped, like lm-deploy, vLLM, unsloth, deepspeed, etc. Awesome.

www.shelpuk.com/post/fine-tu...

One thing you will never think after reading a great book or listening to a great album or seeing a great piece of art is, “I’m really glad this person remained cautious while they were making this and guarded against being perceived as weird.”

Great little list! Thanks!

Agree. We're testing it now and it looks strong.

Or just o1 in the API on release day for its biggest customers would be nice.

Learn something new every day.

Scoop a jar of water out of the ocean and put a lid on it...Study it in its segregated state.

Where is the ocean in that jar?

Where are the tides and the currents?

Pour it back into the ocean and it returns to its integrated state. The temporary entity no longer exists.

- Jed McKenna

If you are new to Bluesky and are an AI Enthusiast.

This hand-picked list (#2) of Newsletter AI writers on Substack, these writers and experts covers a broad spectrum of insights for you that I highly recommend.

Strong in breaking news, LLMs, Generative AI impact, etc.

bsky.app/starter-pack...

The Illogical Logic of Agents: Why They Suck and What We Can Do About It The Eightfold Path to Better Agents

Why we're still in the Atari 2600 era of AI agents and not the PS5 era.

My team works on building agents every day and every team I talk to keeps running into the same problems, whether they have 1M or 10B in the bank.

danieljeffries.substack.com/p/the-illogi...

Playing Atari with Deep Reinforcement Learning We present the first deep learning model to successfully learn control policies directly from high-dimensional sensory input using reinforcement learning. The model is a convolutional neural network, ...

And the paper that started the trend (way back in 2013): arxiv.org/abs/1312.5602

Generative Language Modeling for Automated Theorem Proving We explore the application of transformer-based language models to automated theorem proving. This work is motivated by the possibility that a major limitation of automated theorem provers compared to...

A few more. Published by Ilya in 2020: arxiv.org/abs/2009.03393

I'm looking for arxiv papers on o1 style reasoning. include the original let's... Several recent arXiv papers have explored o1-style reasoning models and their training methods. Here's an overview of some key papers in this area: The...

www.perplexity.ai/search/i-m-l...

Mistral disagrees. And come on, the EU has at least contributed the first convoluted AI legislation in the world. First it regulate, last to innovate!

A reasoning dataset is one that is built to help the model understand how to think through problems with chains of thought reasoning (or other approaches) and special tokens and that can be used with RL to guide the model to reason better through reward/punishment.

Deepseek: The Quiet Giant Leading China’s AI Race Annotated translation of its CEO's deepest interview

Amazing article on DeepSeek, the folks because the excellent open source o1 model, R1.

"While Americans excel at 0-1 innovation, Chinese excel at 1-10 application development."

The Chinese are smashing Americans and Europeans in the open source AI race.

www.chinatalk.media/p/deepseek-c...

It doesn't do quite as well on my private reasoning test as o1 but it is damn good and I suspect if we had reasoning datasets we could augment as a community we would make more progress than the big labs fine tuning it.

Thanks for the kind words.

Why LLMs are Much Smarter than You and Dumber than Your Cat And What We Can Do About It - A Blueprint for AGI

Why LLMs are smarter than you and much dumber than your cat.

danieljeffries.substack.com/p/why-llms-a...

Illogical Logic: Why Agents Are Stupid & What We Can Do About It // Dan Jeffries // Agents in Prod YouTube video by MLOps.community

My latest talk, "The Illogical Logic of Agents, Why the Suck and What We Can Do About It" is now on youtube.

youtu.be/TbnIA0Er5jA?...

Games are reasonable stepping stones as testbeds for AI progress. NetHack and text adventure games hit on modern AI weaknesses. I literally just gave a talk on why we should take Dungeons and Dragons and other role-playing games seriously as AI challenges.

Men fear thought more than they fear anything else-more than ruin, more even than death. It's subversive + revolutionary, destructive + terrible; thought is merciless to privilege, established institutions...thought is...indifferent to authority, careless of the well-tried wisdom of the ages.

- BR

Lol. Awesome.

What are the best machine learning accounts to follow on here? Let me know the ones you love!

Discovered the world's finest cracker today.

Paper thin Italian crackers nicknamed sheet music because they're so thin you can read sheet music through them.

You need these in your life.

Trust me. I'm a doctor.

It's the same fate if you participate. :)