Call to ML-Montrealers: The montreal channel in the Latent Space discord is holding a paperclub Friday Apr 26th 12:00pm at 'Le Petit Dep' (Griffintown).

We'll be covering the 'Towards Efficient Generative LLM' paper. Links below :)

Wow - it's been quite the month.

We announced Devin just a few weeks ago. It's been awesome to see all the support and feedback we've gotten since.

AI agents are just getting started. Two years ago it was unthinkable that an AI could write code, run commands, and browse the…

The upcoming Llama-3-400B+ will mark the watershed moment that the community gains open-weight access to a GPT-4-class model. It will change the calculus for many research efforts and grassroot startups. I pulled the numbers on Claude 3 Opus, GPT-4-2024-04-09, and Gemini.…

🦙 All Llama3 release details & highlights

8B and 70B sizes: SOTA performance on most benchmarks for both Instruct and Pre-trained.

Currently training a 400B+ parameters model which will be released alongside the full Llama3 paper later on:

Created a new human evaluation set…

Huge shoutout to swyx 🥳
I reached out to him based on the quoted tweet

After a few weeks, he directly connected me with the CEO of one of the hottest AI and DevTools startup (raised over $14M) for a DevRel/Evangelist role

Super excited to interview 🤞

ai airchat is popping.

i asked for an explanation of TIES-merging (ft Maxime Labonne’s model merging guide) and nisten obliged:

air.chat/nisten/post/82…

airchat is basically twitter remade in Naval’s image, but one benefit is that you cant write in pompous corpo language,…

TIL about GitHub's upcoming Copilot Workspace that Thomas Dohmke covered today at #TED2024

as both the End of Localhost guy and the AI Engineering guy this is maximum venn diagram overlap for me

also a buuuunch of ai devtools startups are boutta be sherlocked holy shit

following up on some repos from the stream (sorry for delay, merely got lazy)

- yikes github.com/swyxio/devin-d…
- @studentfoundersclub github.com/swyxio/devin-c…
- building a Temporal workflow by reading docs and setting up a local cluster (cant remb who asked for this)…

if you ever have to do any public speaking at all, do your audience and yourself a favor by watching m.youtube.com/watch?v=l9JXH7…

ben has done a ton more worthwhile things since then but if this is your first intro to ben i will have done some good today

Recommended reading

.swyx quote but make it cinema

Our Reka Tech Report / Paper is out! 🔥

Tech reports with completely no information are kinda boring so we’re revealing some interesting information on how we train our series of Reka models including tokens, architecture, data & human evaluation workflows. 😃

We tried…

in the past year:

gpt 4.0: gpt-4-0314
gpt 3.9: gpt-4-0613
gpt 4.3: gpt-4-1106-preview
gpt 4.4: gpt-4-1106-vision-preview
gpt 4.2: gpt-4-0125-preview
gpt 4.5: gpt-4-turbo-2024-04-09

lets be clear what we are talking about when saying another model is 'gpt4 level', because this…