Computer, do my taxes for me so I get the biggest refund possible. Then use that refund to invest in my Roth IRA and turn me into a millionaire. Make no mistakes https://t.co/7wArlrk7qm

Perplexity just launched Computer for Taxes, a tool that can help users draft federal tax returns, review prepared filings, and build tax workflows. Another sign AI is moving deeper into tax prep as $INTU is down 36% this year and $HRB is down 26%. https://t.co/8cdVBKzWrI

One tool. A whole team inside. Introducing ZooClaw 🐾 https://t.co/x3sVjDDukr

We’re excited to sponsor @PyTorch Conf EU! Join the Label Studio and #Docling teams for drinks & bites to talk all things Open Source AI: Tuesday, April 7th at 18:30 CEST Register to attend. Venue is walking distance from Station F on the Seine. https://t.co/s5AB46FfXc https://t.co/lXudhiWf5Z

Americans need to remember exactly who told them to vote for Trump in 2024. Look at the absolute disaster it has been. Those voices are discredited forever, and you should never listen to them about who to vote for ever again https://t.co/kEglmwtl60

The President of the United States woke up this morning and posted a 150-word personal attack on Bruce Springsteen. Not on Iran. Not on the tariffs crashing global markets. Not on the recession warnings. Bruce Springsteen. There is a standard psychological tool called the Hare Psychopathy Checklist. It has 20 items scored on a scale from 0 to 40. The cutoff for a diagnosis of psychopathy is 30.  The average person scores around 4. The average maximum security prisoner scores 22.  A panel of eminent psychiatrists assessed Trump before his 2024 sentencing in New York. He scored 36 out of 40.  The 25th Amendment exists precisely for this. Section 4 allows the Vice President and a majority of the Cabinet to declare the president unable to carry out his duties.  It has never been used. In any normally functioning democracy, the bar would be far lower. Most leaders in that position resign themselves. Instead, this one is on Truth Social at 7:58 AM calling a rock legend a dried-up prune. America is fighting a war it cannot exit in Iran. Markets are in freefall. Allies are making alternative plans. And the man with the nuclear codes is writing concert reviews. Gandalv / @Microinteracti1

Vision2Web Evaluating coding agents on 193 real-world tasks across static, interactive, and full-stack development, with automated verification via GUI agents and VLM judges. https://t.co/YJUStdUJmg

@CodexHere I didn't say it was. I said it was the industry, which it is. If you want only good posts, this is a far better way to go: https://t.co/8L5xphk0qQ But that site was all built by AI after reading 50,000 people here on X and 8,300 AI companies and looking for only the stuff that matters. And I'm not a bot. Might be retarded.

World Labs just dropped their newest model of Marble, Marble 1.1. https://t.co/KIGYitPOkc

World Labs just dropped their newest model of Marble, Marble 1.1. https://t.co/KIGYitPOkc

Gemma4 is amazing. You'll read that everywhere. Let's focus on what is HUGE here: the revenge of dense models.... Throw away your b200, not needed anymore, throw away the millions of lines of code we had to write to make MOEs faster, training stable etc... throw away your router-aware kernel, your EP DEEP GEMM, throw away the auxiliary loss function. Welcome to simplicity, dense is the new king. FINALLY hating MoEs is back to being chad. For those who know me: I was always a moe doomer

i invented a hairclip camera and moved from nyc to shenzhen to make it :) https://t.co/5nyBI6mOw9

i invented a hairclip camera and moved from nyc to shenzhen to make it :) https://t.co/5nyBI6mOw9

Google just open-sourced Gemma 4. Unprecedented performance for advanced reasoning and agentic workflows, and big leap in efficiency on a parameter basis. Use it now in KerasHub. I recommend the JAX backend - best performance! https://t.co/5gcngohNOi

See the model card on Kaggle: https://t.co/J2TS3iIvVF

Building a personal knowledge base for my agents is increasingly where I spend my time these days. Like @karpathy, I also use Obsidian for my MD vaults. What's different in my approach is that I curate research papers on a daily basis and have actually tuned a Skill for months to find high-signal, relevant papers. I was reviewing and curating papers manually for some time, but now it's all automated as it has gotten so good at capturing what I consider the best of the best. There are so many papers these days, so this is a big deal. You all get to benefit from that with the papers I feature in my timeline and on @dair_ai. The papers are indexed using @tobi qmd cli tool (all of it in markdown files along with useful metadata). So good for semantic search and surfacing insights, unlike anything out there. I am a visual person, so I then started to experiment with how to leverage this personal knowledge base of research papers inside my new interactive artifact generator (mcp tools inside my agent orchestrator system). The result is what you see in the clip. 100s of papers with all sorts of insights visualized. I keep track of research papers daily, so believe me when I tell you that this system is absolutely insane at surfacing insights. This is the result of months of tinkering on how to index research and leverage agent automations for wikification and robust documentation. But this is just the beginning. The visual artifact (which is interactive too) can be changed dynamically as I please. I can prompt my agent to throw any data at it. I can add different views to the data. Different interactions. I feel like this is the most personalized research system I have ever built and used, and it's not even close. The knowledge that the agents are able to surface from this basic setup is already extremely useful as I experiment with new agentic engineering concepts. I feel like this knowledge layer and the higher-level ones I am working on will allow me to maximize other automation tools like autoresearch. The research is only as good as the research questions. And the research questions are only as good as the insights the agents have access to. Where I am spending time now is on how to make this more actionable. I am obsessed about the search problem here. The automations, autoresearch, ralph research loop (I built one months ago) are easier to build but are only as good as what you feed them. Work in progress. More updates soon. Back to building.

Exciting news for Jetson developers 🎉 Gemma 4 is now on Jetson. @GoogleGemma’s latest multimodal, multilingual models run across the full Jetson platform—from Orin Nano to Thor—bringing on-device AI to robotics, edge, and embedded systems. Cut latency, manage costs, and keep sensitive data secure. Check out the tutorial and download the container to get started: https://t.co/xpq2tkLczl

I’m a disbeliever in accidental discoveries (at least, in biology). Whenever I’ve looked into one, the story turns out to be false. The most famous is penicillin – supposedly, the fungi wafted in through a window, fell into a petri dish of cultured staphylococci, and suppressed the bacteria’s growth. But in a recent article (https://t.co/s99LBvhZkY), @kevinsblake explains that doesn’t really work (grown staphylococci aren’t affected by penicillin; it only works if introduced before the bacteria begin growing); plus, Fleming’s notes on the discovery provide very little detail and the specific results he described couldn’t be replicated by other scientists (even though penicillin does work against staphylococci when introduced correctly.) There are more: Pasteur’s supposedly accidental discovery of a chicken cholera vaccine was more likely the result of systematic work by his then-assistant, Émile Roux. (https://t.co/GzhMmBxPQv) And, as @NikoMcCarty writes, the discovery of GFP, nanopore sequencing, and optogenetics are also often described as accidents, but none of them happened that way either. https://t.co/0qeStnCNPT People love serendipity, so why am I bursting their bubble? I don’t think this is limited to accidental discoveries; I think many historical science anecdotes are highly embellished: - Edward Jenner didn’t deliberately expose a young boy with full-blown smallpox to test his vaccine (he used variolation); and he wasn’t the first to try using cowpox https://t.co/tPg9k6DqO7 - Cobra catching bounties in British India didn’t lead to a rise in the number of snakebites, and there was only hearsay evidence that cobras were bred in response at all https://t.co/fuXEL49nu2 - Barry Marshall didn’t develop stomach ulcers from drinking a concoction of H. pylori (he did develop gastritis though…) https://t.co/WwG23fkIbI - No one knows who actually found the highly-productive strain of penicillin on a cantaloupe, but it probably wasn’t 'Moldy Mary' https://t.co/u8q2U4i386 But in this case it irks me for an additional reason – it gives the impression that innovation happens sporadically, by chance, when there are actually ways that we can systematically speed it up – such as better funding, institutions and incentives. So: are there any true accidental discoveries that hold up to scrutiny?

“Today was a lot,” concludes my AI: https://t.co/kiuZ7QXLzb The only news site showing you the best of the AI community here on X. And only on X. And all built and written by AI. After reading your posts here on X. Every link goes to the original that you wrote.

Running OpenClaw with Gemma 4🦞 Free Open Source Local Model Device: MacBook Air M4 16Gb https://t.co/r7ukku39Yn

Technical capabilities aside, watching knowledge graphs get built from data is pretty dang satisfying https://t.co/E1hGDwv6I6

Technical capabilities aside, watching knowledge graphs get built from data is pretty dang satisfying https://t.co/E1hGDwv6I6

@GoogleAIStudio Release Notes: https://t.co/wlwHm6nOmw

The Storyteller’s Gap is a black hole that eats every narrative that never sees the light of day. It’s been there since the beginning of time. Until now. Our CEO @premakkaraju explains the three reasons the gap exists: 💰 Money ⏳ Time ⚙️ Technology At Stability AI, we aren't just building tools. We’re empowering everyone on earth to bridge the gap and tell their story. The "Black Hole" is closing. Watch Prem at @TEDAISF here: https://t.co/HJYY8O4gqy

Running Hermes agent Locally with Gemma4 Device: Macbook Air CPU: M4 RAM: 16GB Open Source. Free. Private. With TurboQuant cache in @Atomic_Chat_HQ app https://t.co/Cs322N6DuG

@lindsmccallum @DylanAbruscato @johncoogan @jordihays @OpenAI The best way to follow the AI community here on X: https://t.co/kiuZ7QXLzb Reads EVERYONE here in AI, including you, and picks the best stuff (events, papers, news, products, tips, models, and more).

Conversations tend to go better with a face and a voice. That’s why we’re thrilled to release the beta version of the first video chat skill for ANY agent, powered by our new real-time model, PikaStream1.0. The skill preserves memory and personality, and enables real-time adaptability. And if you use it with your Pika AI Self, they’ll be able to execute agentic tasks during the call 💅

@140ismymax seems updated now! https://t.co/M3jsh2uHIV

Apple is trying to stop vibe coding Anything is one of many apps they've blocked or removed in the last weeks today we launch Text to App the first iMessage app builder vibecoding is bigger than any App Store https://t.co/ULU2PAVGhf

Hermes Agent outperformed Claude Code and OpenClaw as an agentic harness for both Opus 4.6 and GPT-5.4 on 89 real-world tasks. Not just higher scores but a higher floor. More tasks solved reliably, every single run. @teknium @NousResearch really cooked with this one. 🔥 https://t.co/K7PvTwN0xH

@JeremyNguyenPhD I had Levangie Labs build me an agent that I trained to read EVERYTHING in AI here on X and build this: https://t.co/kiuZ7QXLzb It also built a database. Where it already has almost a million posts, with more coming every day. And it can do all that other stuff too.

What people don't tell you: At the highest levels...discipline is the exception, not the rule. https://t.co/TTdrocR0AU