After two years of building solutions that use LLMs heavily I could not keep my mouth shut any more. Why? Because we are being fed fan fiction dressed up as an inevitable future on a daily basis.

So I decided to put some words on a page.

blog.christoolivier.com/p/calling-bu...

The most interesting divide I see in GenAI today is the difference in commentary between those who build with GenAI assistance and those that build GenAI based products. When you know how the GenAI application is made you realise that Context is king and your data engineering must change.

If you are looking for a great soundtrack to get some work done then look at 12 O'Clock Club. You wont be disappointed! (Coding with good vibes, not "vibe coding")

open.spotify.com/artist/6tYSe...

If I hear another tech professional say "reasoning model" or insinuate that LLMs can reason then I am going to scream my head off!! What is wrong with tech people? How many times does it need to be pointed out that these things cannot "think" or "reason"?!! Can your bubble sort algorithm reason?!!

Is anyone else having issues with Google Cloud's Vertex API giving 429 errors for their Gemini Flash 002 models?

I am being stonewalled on questions of capacity and get the feeling that Google is not able to serve demand to PAYG users.

I've spent the last two years scouring all available resources on RLHF specifically and post training broadly. Today, with the help of a totally cracked team, we bring you the fruits of that labor — Tülu 3, an entirely open frontier model post training recipe. We beat Llama 3.1 Instruct.

Thread.

I don’t need to go on social media to have my worldview challenged I am in theoretical physics I have a new existential crisis daily

Snow in London, short days, long evenings—perfect for picking up a new skill. Why not learn #German this winter? ❄️📚 More info here: www.languagecoach.co.uk

Absolutely love this. Very nice write up and something that I agree with. Thank you @principalengineer.com

@languagecoach.co.uk good to see you here! Hope you have been doing well.

There is a lot of energy going into fine-tuning models, but specialized medical AI models lost to their general versions 38% of the time, only won 12%.

Before spending millions on specialized training, might be worth exploring what base models can do with well-designed prompts.

The Onion should buy Elsevier next

Categorically insane to let me do a speech called this

I love that with all the tech companies rushing to add "AI" to their products, and new ones pushing "AI" as a reason to exist... AI is becoming less of a differentiator. Its back to having a reason to exist other than "we have AI!" 😂

👋 from London. I work on Climate Habits, a project to help regular people make lifestyle changes, and I’m enjoying living with green tech (ASHP, EV, repair cafe, reclaimed wood etc) at home

climatehabits.co.uk

Which reminds me - I need to finish off and post part 2 of my guide to artificial neural networks.

Part 1 is here if you're interested:

jonayre.uk/blog/2024/10...