Enhancing RAG with Application-Aware Reasoning Neat trick to improve RAG systems: give it the relevant knowledge and show it how to apply it. Very simple and effective! This approach also works well with AI agents. Pay attention, AI devs. Here are my notes: https://t.co/PxZHcUjc6Q

Announcing Ambient Diffusion Omni — a framework that uses synthetic, low-quality, and out-of-distribution data to improve diffusion models. State-of-the-art ImageNet performance. A strong text-to-image results in just 2 days on 8 GPUs. Filtering ❌ Clever data use ✅ https://t.co/QPZzowf8rN

Top 4 open-source LLM finetuning libraries! From single-GPU “click-to-tune” notebooks to trillion-param clusters, these four libraries cover every LLM finetuning scenario. Understand which one to use, & when...👇 https://t.co/rQ5iYWniNg

btw a shit ton of amazing learning material + open-source code for GPU programming ($150K worth) is linked on the latest @GPU_MODE news post a year ago when I was an undergrad I was scouring the internet for these kinds of resources, plz take advantage of it! https://t.co/e2d5BDNYfX

checks out though I can only comment on Amp @AmpCode let me delete todos please :( https://t.co/xxXIFi0l8F

lol my ChatGPT is not working the way it should https://t.co/hI2e8tPMmx

nice now I got amp to create prs for me with tickets from linear to track progress. slowly building out the pipeline https://t.co/cNw62TGsf8

PerceptionLM: Open-Access Data and Models for Detailed Visual Understanding "In this paper, we study building a Perception Language Model (PLM) in a fully open and reproducible framework for transparent research in image and video understanding. We analyze standard training… https://t.co/E8h5rzyQTP

Scaling Instruction-Tuned LLMs to Million-Token Contexts via Hierarchical Synthetic Data Generation "we introduce a novel post-training synthetic data generation strategy designed to efficiently extend the context window of LLMs while preserving their general task performance.… https://t.co/QB8Afzqum1

One challenge no AI model has been able to do well: "create a coherent, thematic puzzle for a D&D game. The puzzle should be challenging, but solvable" The current big models are much more on theme than older ones, but still are either too easy or hard (And love similar puzzles) https://t.co/NaL0xCQ13T

+1 more

“If evals is just a metric, then you’re thinking about evals wrong. It’s not a metric, it’s a entire process.” @HamelHusain joined our Building with LLMs course to talk about why most teams get AI evaluation wrong, and what it actually takes to improve AI products. The hardest… https://t.co/LKkWvR6B3c

Alex Wang timed his exit perfectly. Scale AI was being beaten by a startup rival that never raised VC and was profitable the whole time https://t.co/b5uhKQtyN3

i must say it's kinda impressive that AI paper discussions get Community Noted, this is very good for constructive scientific discussion imo https://t.co/3V7kPpKzXF

We've been negotiating a $2M contract to get AMD on MLPerf, but one of the sticking points has been confidentiality. Perhaps posting the deliverables on X will help legal to get in the spirit of open source! https://t.co/cnOiumwmHl

CVPR 2025 papers pt. 3 - EdgeTAM EdgeTAM lets you run high-quality video object segmentation and tracking at up to 16 FPS right on an iPhone 15 Pro Max more papers: https://t.co/1VlLn2BWxl ↓ more https://t.co/CeXzenugsP

We found it surprising that training GPT-4o to write insecure code triggers broad misalignment, so we studied it more We find that emergent misalignment: - happens during reinforcement learning - is controlled by “misaligned persona” features - can be detected and mitigated 🧵: https://t.co/BW6YCnf3oE

Many of the founders in my network are exclusively hiring former founders. Especially to fill marketing, operations, growth, and chief of staff roles. Seems to be a growing trend. https://t.co/Lt4RVtgcKt

It feels like a steal. An average AI or RAG course costs $2-3K. And an AI Engineer at Microsoft earns $375,000 But now, some of the top RAG and AI experts are running a 5-session live Zoom masterclass and anyone can join for free. Upcoming sessions: 1. Wed, Jun 25 - I don't… https://t.co/y0TnRVRzVj

Leaky Thoughts Hey AI devs, be careful how you prompt reasoning models. This work shows that reasoning traces frequently contain sensitive user data. More of my notes below: https://t.co/y0PVfqeKCw

slow year... I'm getting cooked by @HamelHusain, absolutely COOKED https://t.co/GZfEXtwl4h

Does MCP Kill (Centralized) Vector Search? In the post-MCP world, AI agents will interact with the external world through MCP tools. There is a very valid question on whether you would want to have a centralized search index at all if agents can just directly interface with the… https://t.co/oEtREh6UVx

Making touch sensors has never been easier! Excited to present eFlesh, a 3D printable tactile sensor that aims to democratize robotic touch. All you need to make your own eFlesh is a 3D printer, some magnets and a magnetometer. See thread 👇and visit https://t.co/p3eKwLnCtF https://t.co/Njq7JYzNJc

I get tons of email spam but buried in the noise, there’s real signal. Forward lets people pay to reach your inbox, so the important stuff stands out. Impressive how fast @pol_avec was able to bring this idea to live now that stablecoins are going mainstream @CoinbaseDev https://t.co/j0TFRQEsX2

We present an Autoregressive U-Net that incorporates tokenization inside the model, pooling raw bytes into words then word-groups. AU-Net focuses most of its compute on building latent vectors that correspond to larger units of meaning. Joint work with @byoubii 1/8 https://t.co/QqQ9qb4Xfv

Midjourney now does video, and, like Midjourney itself, its advantage is that it has features that allow you to create styles that are hard achieve with other video creation tools & feel less like standard video pastiche Here are a bunch of five second clips I made, for example. https://t.co/AObNX84dGI

Workaccount2 on Hacker News just coined the term "context rot" to describe the thing where the quality of an LLM conversation drops as the context fills up with accumulated distractions and dead ends https://t.co/2oWaMhlZDi https://t.co/8kZgHNHHG0

My friend @JoeEHoover is hiring evals folks - reach out to him directly via the email in the post if interested https://t.co/a5aPNALADz

Gemini is good at processing video (using frequent screenshots & audio transcripts). I gave Gemini a video on a historical recipe, it was able to find visual elements not mentioned in the transcript. It is not hallucination-free, but there are lots of new use cases for screening https://t.co/zg1l3fdwR6

+2 more

I got a cease and desist from DocuSign for my free SaaS. A couple of months ago, I saw a tweet from @awilkinson: “I just found out how much we pay for DocuSign and my jaw dropped. What's the best alternative?” Me being naive, I thought “how hard could would it actually be to… https://t.co/riURobOC3A

I enjoyed this section in the LLM evals course reader. Btw 35% discount here: https://t.co/3botO05IlE It suggests three gulfs builders have to navigate when building AI systems. First, the Gulf of Comprehension. Don't assume we know our data; that's a trap. Real-world data is… https://t.co/6E8ngo05Do

PSA https://t.co/9PHWrynbMj

feeling like a ceo, assigning two teams to the same task, firing the one that fails https://t.co/HmXQ4NjFas