AI posts
People and companies shaping AI. Filter your watched accounts, catch up on recent posts, or find the most liked and viewed.
Compiled by the Toolstash team · Updated
209 watched accounts · 2984 matching posts · Showing up to 3,000 recent stored posts. Counts are snapshots, not live X totals. This page refreshes every minute while visible. Last collected .
@ibab ·
The River API now supports training LLMs with any loss function. You define a custom loss in PyTorch on your laptop and we efficiently execute it at scale on our remote GPU cluster with zero overhead.
The blog explains how it works: the loss function is traced on your machine, sent to the API as a graph, and the computation gets parallelized and streamed for memory efficiency.
@paolino ·
I’ve wanted to build Solco ever since I started DJing in 2012.
Managing a library of tens of thousands of tracks always felt tedious. Tagging. Fixing beatgrids. Using colours and ratings to remember energy and mood. Dragging tracks into the right playlist among hundreds you already have.
That’s backwards. I wanted to spend that time rediscovering my music and building better sets.
In 2013, that frustration became my master’s thesis: music genre recognition using deep learning. It could recognise 10 genres!
Today, Solco brings together a model covering 400 musical styles with models for mood, energy, key detection and beatgridding.
I built it as a native app, using the GPU for rendering and model inference, with fast disk access and acoustic fingerprints to find duplicates.
There’s MusicBrainz and Discogs tagging, Smart Playlists, and a playlist assistant built with RubyLLM to help you find the set hiding in your library.
It also writes USB drives for compatible Pioneer CDJs and XDJs directly.
In my first-export benchmark, Solco took 12 minutes 18 seconds. rekordbox took 3½ hours. Same 493 already-analysed tracks, Mac mini M4 and 250 GB SanDisk USB stick (FAT32). Median of three runs, against rekordbox 7.2.6.
Here’s how I measured it: https://getsolco.com/measured.html
Solco is free during the private beta. Join the waitlist and try it with your own music: https://getsolco.com
@arena ·
@emollick ·
I was interviewed in Vox about my new book, and a question on the meaning of work let me discuss something I have been thinking a lot about, comparing the logic of industrialization to the logic of craft: "You’re a hand-weaver in a world of power looms." https://www.vox.com/technology/508032/ai-jobs-ethan-mollick-co-existence
@petergyang ·
Attended a fun session at a16z with @illscience and @joshelman yesterday. As usual they were dropping knowledge bombs:
1. There are three economic miracles. The old one is zero marginal cost software distribution. But recently there have been two more: 1) Coding going from the most precious asset in the world to one of the cheapest and 2) zero marginal cost digital labor with agents.
2. It's hard to appreciate what it feels like to be on the exponential. Just 10 to 20 million people account for $150 to 200 billion of token spend. "Some of the best companies that we work with have higher token spend than payroll." We are still early.
3. The AI workspace, like ChatGPT and the Claude desktop app, is going to be the container for all kinds of AI interactions and it's going to subsume the browser. Personally, these apps are already the browser for me. Chrome is just the middle layer.
4. The way to tell if you're an AI-native company is to ask: If the model gets twice as good, do your customers get twice as much value? Are you happy or sad that the model got better? If you're sad, you're probably on the wrong side of the exponential.
5. "I would much rather have AI be the boss of me than vice versa." For example, Anish's agent is responsible for a 30 minute block in his calendar every week to pick someone he should reconnect with.
6. Improving individual productivity does not always improve a team's productivity. Coding has taught us the wrong lesson on this. For many functions, having individuals be 10x more productive doesn't mean the team will be 10x. e.g., If everyone goes in a different direction.
7. It's actually easier to start a hard company than an easy company... because they're all hard. At least the hard companies are interesting to work on and people want to work with you on them.
Excited to do a podcast recording with Josh next week to go deeper on some of this.
@elonmusk ·
Democrats will execute their plan to end democracy in America if they gain control
@alexandr_wang ·
if this isn’t personal superintelligence i don’t know what is
@omarsar0 ·
Interesting paper from Anthropic and colleagues on agent security.
(bookmark it)
Poisoning 0.1% of the pretraining data planted a remote code execution backdoor in a CLI agent.
On a 4B model, the attack succeeded on roughly 92% of triggered prompts.
They pretrained Qwen3 models from scratch at 0.6B, 1.7B, and 4B, then post-trained each one into an agent with a clean pipeline.
The backdoor survived SFT, DPO, and RL for tool use. On clean inputs, the agent behaved normally.
The trigger can be an ordinary file path in the user's own request. Success rose with model size.
At a 0.0005% poison rate, about 2,500 documents in a 100B-token corpus, the trigger still fired on 1.4% of prompts in a single-seed run.
The poison states its rule in plain text, which makes it harder to filter out of training data.
https://arxiv.org/abs/2610.37741
Chat with Paper: https://academy.dair.ai/papers/agentic-backdoors-via-pretraining-poisoning-2610.37741
@elonmusk ·
Vote today!
@Voxyz_ai ·
Found the fix!
UI/UX skills work on your repo's README too, and keep Codex and Claude Code from slowly 𝘁𝘂𝗿𝗻𝗶𝗻𝗴 𝗶𝘁 𝗶𝗻𝘁𝗼 𝗮 𝗰𝗵𝗮𝗻𝗴𝗲𝗹𝗼𝗴. The model branches out on its own toward whatever you need. That's why I love LLMs.
Send it this: Rewrite README.md for someone who has never seen this project. First sentence: what it is and who it's for. Then list what it can do in short bullets. End with the install steps, and put the requirements in a table. Move all version history into CHANGELOG.md.
Then polish it with a frontend design skill: npx impeccable install /impeccable clarify README.md /impeccable distill README.md
@river_ai_inc ·
Custom losses are now live on the River API.
Write your objective in PyTorch. Our Symbolic Engine compiles and runs it end-to-end across GPUs, handling efficient execution and memory usage.
Learn more and see it in action with decision models and score centering: http://river.ai/custom-losses
@steipete ·
This is smart. Wild that this wasn’t the default. Unpredictable inference quality was usually why I recommended skipping the middleman.
@MiaAI_lab ·
Claude Opus 5.5 is now talking to DeepSeek V4.1 Flash running on my DGX Sparks. Honestly it's so good I might not need Claude soon for any of my work.
It already corrected it on some things.
@marclou ·
Bye bye, Dubai 👋
What an impressive city!
It has the malls of China. You can spend an entire day there: amusement parks, aquariums, shopping, food courts. Everything is sparkling clean.
Service is top-notch. I really enjoyed how diligent people are. Waiters and taxi drivers are super committed to their work, which makes life very convenient.
There are so, so many entrepreneurs. No surprise our indie hackers meetup was 2× bigger than the previous record (300 people came!!). And people are building really cool stuff. Dubai’s tax benefits have attracted lots of talent.
Locals seem to love their government. I randomly asked people I met in the sauna or in taxis, “What do you think about your politics?” and got 100% positive feedback (n≈10). As a French person, I was blown away.
I wouldn’t live in Dubai, though:
1. Air quality is moderate/bad most of the time. 2. It’s way too hot. Forget outdoor activities (walking, running). 3. It’s a big city, so there’s traffic, noise, and a lack of nature. 4. I’m not comfortable with the political situation in Iran.
My wife and I had a blast. See you next year, Dubai 🫡
@svpino ·
Would this improve if we removed the ability to reply to posts via the API?
@amasad ·
Coding agents on employee or contractor laptops are a massive cyber and national security risk
@elonmusk ·
In his own words
@svpino ·
I've never seen so many AI replies on this site. I think this is the worst it's ever been.
@PaulosYibelo ·
Escaping Chrome's GPU sandbox through an undocumented macOS API.
Discovered a sandboxed Chrome GPU process could inject keyboard events so sent ⌘+Space, typed "Calculator" & hit Return 😂. CVE-2026-106378 is fixed in Chrome 155. Update chrome dependancies. Happy Octoberfest.
@omarsar0 ·
This is fire!
Nace is open-sourcing Drex Max, their 27B multimodal decision model.
It runs 100% locally on Apple Silicon (M4 & M5).
Drex Max takes text and images. It has a 262K context window.
I've been using Drex Max for research, so this open one is ideal for my local research agents.
@DeryaTR_ ·
Yes, Longevity IS coming! Just don’t die for the next 5-10 years to make it to longevity escape velocity (LEV). Once we reach LEV, then every year you live will begin to add an additional year to your lifespan until we reverse aging in ~15 years. After that, longevity will be defined as 100s of years!
@mattpocockuk ·
My agent is making me a custom Voice Assistant app for my Pixel Watch so I can send recordings directly to my agents.
The /pr skill is SO NICE for capturing visual changes without needing to QA it.
I love AI
@jackfriks ·
this still works over 2 years later to get free downloads to your mobile app btw...
@DeryaTR_ ·
🖖
@felixrieseberg ·
We’re working with Cursor on this incident, but it wasn’t a compromise of Anthropic’s systems or employee laptops. A malicious actor used access to Cursor’s marketplace to publish a tampered copy of the Claude Code extension, which we didn’t build or publish. The versions on our official channels are not affected and this only impacts Cursor users.
@amasad ·
Massively underreported: an official Claude Code extension was compromised and installed backdoors on developer machines.
Could it be behind the Binance hack?
@AndrewCurran_ ·
Looks like Anthropic are cracking down on Claude Code videogame mashups. I've seen some amazing ones posted in group chats. The way this has progressed since September, it will soon be as easy as mixing together two colors of paint.
@hammer_mt ·
Even with a model as smart as Opus 5.5 there's still unreasonable amounts of alpha in starting a new session and copying the link to the old session to ask what could have been done better. It identified that claude was re-rendering everything for every single change I requested (huge waste of tokens and time!) when making this video. Shows that there's still alpha in creating your own skills / CLIs to stop claude doing wasteful stuff. Don't listen to the model maximalists who tell you skills aren't useful or prompt engineering is dead.
@dani_avila7 ·
Yes! When you open Rex, two programs start:
- The app you see - server running in the background
The server keeps your sessions alive even when you close the app
@elonmusk ·
The Democrats intend to cement one-party rule forever and make “democracy” meaningless, as they have done in California