🔥The 9th Round of Easy Loan, Earn $40 Reward is in progress❗️ ⏰ Promotion Period: January 15th - Feburary 15th, 2025 👉 Register now and check more details at gate.io/campaigns/358

dspy is the only real way to building with smaller models which also include open source ones. it's insurance against that one day they decide to turn off access.

Here's your reminder that you will benefit from using dspy even if you dont plan to evaluate and optimize your prompts. Thanks to DSPy, I've developed a clear mental model for organizing information effectively. Building an agent (often referred to as AI software or programs)

🎙️ Multi-vector retrieval takes the spotlight. Antoine Chaffin, R&D ML Engineer at LightOn, will join Hamel Husain and Shreya Shankar to showcase how multi-vector architectures can overcome traditional single vector retrieval limitations. Don’t miss this deep dive into LightOn’s

"It's like there's a certain minimum number of days, weeks, or months (for different tasks) that you need to spend in play mode just collecting signals, accumulating them, and digesting them. And digesting them just takes a certain amount of time, after which you just come out

Overemphasis on simplicity gives you a big bag of tricks that don’t add up. Parsimony is a better goal. Find the smallest number of pieces that cohesively fix a whole range of problems. Put differently, you need to invent Unix first before it makes sense to create lots of small

the coolest thing about llm's is that they're just text in -> text out, adding a lite structure on top of input fields -> output fields exponentially increasing the ability to compose these together into systems of llms this is the key to why dspy matters.

Calling learning natural-language rules “not real learning” is so backwards. Interacting with an environment to generate abstract hypotheses and turn them into actionable natural-language rules is as “learning” as the word’s natural connotations get. Though gradient-based

Casper Hansen For modern-looking approaches, this was a quick-n-dirty 2023 thread that applies rejection-finetuning from outcome rewards for multi-hop behavior in a few LoCs. This is the ‘RLVR’ paradigm of OAI’s DeepResearch, but replace offline fBC with online PPO. x.com/lateinteractio…

about to open-source "Robofactor" (i apologize for the name 😭 i don't like coming up with names)! Robofactor even includes a DSPy program to write the README lol

There's 8 hours left before our CBRNE track ends (June 19th, 11:59 PM EST) and we award the $65,000 prize pool. A huge thank you to our partners at OpenAI for making this track possible!

I'm looking for 2-3 emergency reviewers for ARR. Papers related to multi-hop search and LLM agents / systems. If you can review in 2-3 days or slightly longer, DM me or reply/email etc.

🤯What we know about RL for reasoning might not hold outside math and code? We revisit established findings on RL for LLM reasoning on six domains (Math, Code, Science, Logic, Simulation, Tabular) and found that previous conclusions drawn on math and code are surprisingly

Late to this game, but it is insanely easy to setup slack webhooks so you can notify yourself via slack when things happen. All done through a single CURL I used to send automated texts or emails to myself when training jobs are done but slack is way better