Manuel

defai maxi | 100% $SOL | ai engineer. all views my own!

[Manuel @manuelsvard]

@launchcoin great tech

[Turing Post @TheTuringPost]

The freshest research papers of the week:

Our top 10:
▪️ Flow-GRPO
▪️ Unified Multimodal Chain-of-Thought Reward Model through Reinforcement Fine-Tuning
▪️ RM-R1
▪️ Scalable Chain of Thoughts via Elastic Reasoning
▪️ X-Reasoner
▪️ Practical Efficiency of Muon for Pretraining
▪️ Grokking in the Wild
▪️ Teaching Models to Understand (but not Generate) High-risk Data
▪️ LLM-Independent Adaptive RAG
▪️ Perception, Reason, Think, and Plan: A Survey on Large Multimodal Reasoning Models

Link to the full list of research in the end of🧵

[AK @_akhaliq]

Omni-R1

Do You Really Need Audio to Fine-Tune Your Audio LLM?

[Thomas Wolf @Thom_Wolf]

🎠 Low Tech AI

[Avi Chawla @_avichawla]

A drag-and-drop visual tool to build AI agents.

Langflow lets you build and deploy AI-powered agents and workflows. Supports all major LLMs, vector DBs, etc.

100% open-source with 61k+ stars!

[Tanishq Mathew Abraham, Ph.D. @iScienceLuvr]

Salesforce introduces:

BLIP3-o: A Family of Fully Open Unified Multimodal
Models—Architecture, Training and Dataset

"we introduce a novel approach that employs a diffusion transformer to generate semantically rich CLIP image features, in contrast to conventional VAE-based representations"

"we demonstrate that a sequential pretraining strategy for unified models—first training on image understanding and subsequently on image generation"

[Alexander Novikov @SashaVNovikov]

After 1.5 years of work, I'm so excited to announce AlphaEvolve – our new LLM + evolution agent!
Learn more in the blog post: https://t.co/UwbM3jjN4t
White paper PDF: https://t.co/KpZUHAZeFm
(1/2)

[AK @_akhaliq]

AM-Thinking-v1 just dropped on Hugging Face

Advancing the Frontier of Reasoning at 32B Scale

[basedfk @basedfk]

I'm paying $200/month for ChatGPT Pro so that you don't have to.

Q: At what stage is cryptocurrency adoption today compared to the historical timeline of internet adoption, particularly in relation to the dot-com boom era?

A: If the internet’s “dot-com boom” was the late ’90s, then crypto is somewhere around that mid-to-late-’90s stage in its development. The technology is still maturing, speculation is running hot and cold, institutional involvement is ramping up, and we haven’t yet seen the true “killer apps” that will pull in the next billion users. We’re likely still in the early innings.

Takeaway: We'll likely see a huge parabolic move, a blow off top within the next 4 years.

[Cognition @cognition]

Turns out Devin's been doing the work of 10 engineers at @BiltRewards (according to their own calculations).

They're using Devin for everything from turning Figma designs into code to investigating GCP issues.

How Bilt is merging 117 more PRs/week with Devin 🔗 👇

[clem 🤗 @ClementDelangue]

8x faster/cheaper @openai Whisper API thanks to Hugging Face Inference Endpoints & @vllm_project!

[Xuan-Son Nguyen @ngxson]

Firefox is open-source on Github, and they experimented with @ggml_org llama.cpp in WASM 👀

Wondering what they are cooking 🧑‍🍳

[AK @_akhaliq]

Bytedance just dropped Seed1.5-VL on Hugging Face

Achieves top performance with a relatively modest architecture, 532M vision encoder & 20B active parameter MoE LLM.

Delivers State-of-the-Art results on 38 out of 60 public VLM benchmarks, demonstrating broad competence.

[David Valerio Gilmore @davidvgilmore]

Prompt-to-MCP server is now live in @workshopai! Create an MCP server and deploy it to @Netlify in 10 minutes or less.

In the below example, I created + deployed an MCP server to wrap the Hacker News API.

Stay tuned … we’re cooking some more MCP features in the next days and weeks.

[AK @_akhaliq]

ByteDance just dropped DreamO on Hugging Face

A Unified Framework for Image Customization

With a single model, DreamO supports ID, IP, Try-On, and Style tasks, and even allows multi-condition inputs. It's lightweight, performant, and achieves state-of-the-art results across these tasks.

[célina @hanouticelina]

🚀 Last week, we've dropped a new release of the 𝚑𝚞𝚐𝚐𝚒𝚗𝚐𝚏𝚊𝚌𝚎_𝚑𝚞𝚋 Python library! v0.31.0

$ pip install -U huggingface_hub

This one’s packed with brand new features for Inference Providers — including LoRAs support, auto mode for provider selection, and embeddings support 🔥🧵

[Qwen @Alibaba_Qwen]

We’re officially releasing the quantized models of Qwen3 today!

Now you can deploy Qwen3 via Ollama, LM Studio, SGLang, and vLLM — choose from multiple formats including GGUF, AWQ, and GPTQ for easy local deployment.

Find all models in the Qwen3 collection on Hugging Face and ModelSope.
Hugging Face：https://t.co/V1WxhQ0fad
ModelScope：https://t.co/Z9Z37FODVN

📷 For more usage examples, check out the image below!

[OpenAI @OpenAI]

Evaluations are essential to understanding how models perform in health settings.

HealthBench is a new evaluation benchmark, developed with input from 250+ physicians from around the world, now available in our GitHub repository.

https://t.co/s7tUTUu5d3

[Julien Chaumond @julien_c]

llama.cpp is now fully compatible with VLMs 💥

HUGE kudos to @ngxson from HF and to the @ggml_org team 💟

Here are a selection of pre-quantized models, ready to be used, from:
- @GoogleDeepMind Gemma
- @MistralAI Pixtral
- @Alibaba_Qwen VL
- @huggingface SmolVLM

Give them a try and share your feedback!

[basedfk @basedfk]

Imagine @base had a Pumpfun that's less than 2 months old

and you can get a piece of it at under 2M MC with revenue sharing

True launchpad ownership and even Pump took 3 months to really start getting traction

What's the ticker?