Yesterday, an engineer asked me how DSPy's MIPROv2 optimizer works. He'd read the code but found it overwhelming. Referring him to the paper helped, but he still lacked an intuition for MIPRO. I then remembered this series of cool animations by Michael. Made all the difference!
MIPROv2, our new state-of-the-art optimizer for LM programs, is live in DSPy @stanfordnlp! It's even faster, cheaper, and more accurate than MIPRO. MIPROv2 proposes instructions, bootstraps demonstrations, and optimizes combinations. Let’s dive into a visual 🧵of how it works!
Announcing Surya Table Recognition! It uses a new architecture to outperform table transformer, the current SoTA open source model. - Recognizes table rows, columns, and cells - Works with complex layouts and rotated tables - Supports any language - Runs locally
A prompt that helps Claude 3.5 Sonnet beat OpenAI's o1 model in reasoning!
Can @AnthropicAI Claude 3.5 sonnet outperform @OpenAI o1 in reasoning? Combining Dynamic Chain of Thoughts, reflection, and verbal reinforcement, existing LLMs like Claude 3.5 Sonnet can be prompted to increase test-time compute and match reasoning strong models like OpenAI o1.…
localhost:5173 needs a new house! 🏠👨💻 My beautiful journey as a developer @ISC_money has come to an end. Grateful for the experience and excited for what's next! Thread incoming on my ISC adventure 🧵🪡 #CareerMoves #DeveloperLife #ISC #CryptoJobs (1/7)
How structured outputs work under the hood (via breakout at OpenAI DevDay) Guess why the first structured output request is slow, but the 2nd+ is fast? Engineering: * Unconstrained token decoding isn't good. The model could pick any token. * Limiting which tokens can be…
Introducing the commit0 interactive environment for coding agents. Challenge: generate Python libraries from scratch. Commit0 is designed with interactivity, dependencies, and specifications as first-class considerations. We include a benchmark with 50+ challenging libraries.
More data makes RAG applications worse, not better. Relying on vector similarity search doesn't scale, and most people aren't talking about this. This is counterintuitive, but experiments don't lie: As you add more documents to a vector-based RAG system, its retrieval…
A goldmine of tutorials about Generative AI Agents! You'll find anything Agents-related in this repository. From simple explanations to the most advanced topics. Star this repo from @NirDiamantAI: github.com/NirDiamant/Gen… The content is organized in the following categories:…
YAML >> JSON for LLM input. Interesting read betterprogramming.pub/yaml-vs-json-w…
Dear Indian founders incorporating in the US, Half a dozen people have now pinged to ask me about this. So I'm posting here to reduce the number of phone calls. here's collective wisdom of a few hundred YC India startups ok the correct company structure. h/t to Anand of @inklehq…
We recently worked with @OpenAI to fine-tune gpt-4o to build the SOTA model for static analysis eval. All the code and data on how they did it is available on their GitHub - github.com/openai/build-h… More details on the static analysis eval benchmark are available on…
Hey AI Devs!! Going to integrate semantic cache on production this weekend. Gathering insights on GPTcache, Redis semantic cache and canonical AI sem cache. Let me know your thoughts
Clip of real time conversation with GPT4-o running on ChatGPT app NEW: Instead of just turning SPEECH to text, GPT-4o can also understand and label other features of audio, like BREATHING and EMOTION. Not sure how this is expressed in the model response. #openai
I am diving into LLM fine-tuning. There is a lack of deep tech content on fine-tuning: • how it works • why it works • what it does to an LLM, etc. There is a ton of high-level stuff, however. I want to grok the first principles of fine-tuning. If you have an excellent…
How do I detect when someone is done speaking when transcribing with something like Whisper? cc @batwood011
Unfortunately, Large Language Models will not consistently follow the instructions that you give them. This is a massive problem when you are building AI systems that require a particular type of output from the previous step to feed into the next one! For example, imagine you…
United States Trends
- 1. Cowboys 71,1 B posts
- 2. soobin 87,8 B posts
- 3. Clippers 11,8 B posts
- 4. KADOKAWA 31,8 B posts
- 5. Norman Powell 2.018 posts
- 6. Jerry 42,7 B posts
- 7. Texans 53,2 B posts
- 8. Mike McCarthy 3.301 posts
- 9. Cooper Rush 11,9 B posts
- 10. sabrina 111 B posts
- 11. Fultz 1.375 posts
- 12. Lindy 3.719 posts
- 13. #EeveeDay N/A
- 14. Eliza 12,1 B posts
- 15. Dyson Daniels 2.868 posts
- 16. #AskShadow 7.440 posts
- 17. Eevee Evolution N/A
- 18. Joe Mixon 10,7 B posts
- 19. Keon Ellis 1.731 posts
- 20. Herta 43,3 B posts
Something went wrong.
Something went wrong.