Walker Bytes @wb_bitsbytes

Presenting this work today at @CVPR from 11:45 AM – 1:45 PM, Poster 451! Swing by if interested, would love to chat! #CVPR2026

chuyi shang @chuyishang

3 days ago

What if the best visual reasoning steps are ones humans can’t specify? 🤔 Existing VLM reasoning is often constrained by language, pixels, and human-designed intermediates. We introduce Latent Implicit Visual Reasoning, where we show that VLMs can discover the best visual

1 1 20 5K 7

Two guys in their bedrooms in Serbia and Poland, English not even their first language built an AI writing app in two weeks and took it from $0 to $100K in three months all from one TikTok account. They tried paying influencers first and got views with zero buyers so they filmed themselves instead and landed their first sale from the very first video. The playbook never changed: find what's already viral in your niche, copy it one-to-one post one video a day, and only get creative once it works. 9 million views, 250K signups, 80% visit-to-signup $100K in three months on a server that cost $35 a month. Their whole edge was one boring rule: don't invent a format, just replicate what's already winning.

Ridark @ridark_eth

17 hours ago

x.com/i/article/2063…

25 12 121 72K 250

@arekfurt @thegrugq this is an obvious consequence of imposing rules about not reporting known breaches instead of imposing liability, period. If companies don't want liability for data that leaks, they need to either ensure security, or not hold the data! x.com/arekfurt/statu…

Brian in Pittsburgh @arekfurt

a day ago

Quite a revelation this week. In 2000, the former IBM VP of Threat Intelligence filed a lawsuit against IBM and AT&T essentially alleging IBM's poor security monitoring was intentional. To avoid understanding the scope of breaches and data exfiltration and having to report them.

keyboard_arrow_left Previous keyboard_arrow_right Next

3 16 52 8K 15

@magnushambleton Not usefully EV positive to say to invest at a stage where companies aren't taking investments. (By the time they are successful enough to have gotten major investors and don't need capital, they have enough momentum to raise additional capital at a much higher valuation.)

Dave Fairtex's latest Market Update is out: Strong payrolls, surging dollar, and a brutal plunge in metals & equities. #economy peakprosperity.pulse.ly/0cxu4ohyzf

A famous math problem that stumped humans for the better part of a century has finally been toppled—by AI. on.wsj.com/4uR7aZJ

Could the US government actually do a one time 50% tax of the AI labs? We asked @hamandcheese, chief economist at @joinFAI, whether Bernie's proposal has any legal path forward. "There are degrees of nationalization on some level every company in America is nationalized insofar as they pay corporate income taxes and abide by our laws." "Bernie's proposal is to take a one-time tax for 50% of OpenAI and Anthropic... very unlikely to ever happen." "The Constitution has a takings clause, you have to pay fair market value for that equity. Unless they're planning to print a trillion-dollar coin, that's not in the cards." "Taking a small non-controlling stake, not the craziest idea in the world. I just think we don't have the political economy of Norway."

Samuel Hammond 🦉 @hamandcheese

3 days ago

Nationalizing AI companies is a terrible idea. But if it happens, it's likely to be done slowly and stealthily through procurement rules and other forms of quasi-regulatory co-optation. This btw is another reason to favor independent verification orgs over direct govt control.

6 6 65 16K 16

Anthropic pays $700K+ a year for engineers who build LLM architectures from scratch. This free 1-hour Stanford lecture walks the exact stack they pay for: attention variants, normalization, positional encoding. The catch nobody says: architecture is the part that matters least. Every lab uses the same published blocks. If architecture were the moat, everyone would have GPT-4. The $750K skill is data, scaling, and systems on top of it. 3 tiers, not 1: Tier 1, from scratch. This lecture. Millions and a lab. Tier 2, fine-tune an open model. 1 night, $16, the skill paying $195K-$350K. Tier 3, no-code on Claude. $0, shipping tonight. Bookmark the lecture. Watch it today. Then read the article below for the tier you can start in.

Moysei @0xMoysei

2 days ago

x.com/i/article/2063…

5 2 57 47K 129

As U.S.-Iran talks show little sign of progress, commercial traffic through the Strait of Hormuz remains sharply reduced, raising concerns about global energy markets and supply chains. @GeoffRBennett speaks to Daniel Yergin, vice chairman of S&P Global, about what a prolonged disruption might mean.

Transformer co-author, @lukaszkaiser, on why he prefers Codex:

Jacob Effron @jacobeffron

5 days ago

I sat down with @lukaszkaiser to get into whether the architecture he helped invent is actually enough, and what's next in generalization, coding agents, RL and more. Lukasz co-authored "Attention Is All You Need," the paper that introduced the transformer and worked on reasoning

3 13 136 60K 183

ColBERTSaR: Sparsified ColBERT Index via Product Quantization @EYangTW et al. present an embedding quantization method that turns a ColBERT index into a true inverted index, yielding an index 50-70% smaller than 1-bit PLAID. 📝 arxiv.org/abs/2606.05568 👨🏽‍💻 github.com/hltcoe/ColBERT…

Marc Davis is a master craftsman. People send him their classic cars. He rips out the engines. 3D scans the cars and then gets to work finding a way to turn these old beauties into modern, all-electric gems. The process costs $75k to $175k. His company - Moment Motor - has done tons of cars and trucks, and, in this episode alone, we take a peek at its work on a 1972 Chevy Blazer (running a Tesla Model S drivetrain), a 1968 Ford Mustang fastback (a Bullitt replica), a Series 1 Lotus Esprit (the James Bond car), a 1970 Mercedes-Benz 280 SL, and a 1960 Ford F-250. youtube.com/watch?v=F0o11h…

Either Sam Altman reads The Pragmatic Engineer (I wrote about this in detail exactly one week ago for paid subscribers) or I'm just a week ahead to what he also sees!

Roger @rdd147

4 days ago

🚨 Sam Altman warns OpenAi and Anthropic are experiencing severe pullback on Ai spending as companies put significant restraints on spending to restrict costs. The company warns investors it’s the first time this has happened in Ai and something we never expected. The buildout

265 787 6K 1.9M 2K

My work at Elicit is out! We spent a lot of time trying to build the world's best systematic review evaluations. I might be wrong, but I think this is the largest AI assisted SLR dataset in the *world* by 10x!!* We benchmark against ~994 papers versus ~100 for the next largest

Elicit @elicitorg

a week ago

How well does Elicit's semantic search work for systematic reviews? Cochrane reviews are the gold standard for evidence synthesis in health and medicine. We sampled 888 reviews across 12 MeSH areas. For each Cochrane systematic review, we ran Elicit using only the review title

2 1 19 8K 4

Matt's guide is great but this line is pretty funny: "They're so lit up by being able to build anything that they don't do anything else. Then they launch and there are no users." 😆

Matt Van Horn @mvanhorn

5 days ago

x.com/i/article/2061…

122 367 3K 925K 9K

Slipstream: Locality-Aware Graph Index Construction for Streaming Approximate Nearest Neighbor Search Speeds up insertions in graph indexes for streaming nearest neighbor search by reusing candidates from previous insertions 📝 arxiv.org/abs/2606.02992 👨🏽‍💻 github.com/suey3141/Slips…

Residual Primitive Fitting of 3D Shapes with SuperFrusta Another great work done by our intern Aditya Ganeshan @AdobeResearch. We introduce a framework for converting 3D shapes into compact and editable assemblies of analytic primitives.

@heyitsalexsu @Discoplomacy - It's $300k plus equity - so it's easily in the millions, maybe even paying better than top law firms pay partners. x.com/heyitsalexsu/s…

Alex Su @heyitsalexsu

6 days ago

Anthropic is looking to hire a law professor and pay them 300k a year, it’s gonna be as competitive as a scotus clerkship

keyboard_arrow_left Previous keyboard_arrow_right Next

19 64 898 131K 283

the technology worked but not in a way that provided value, also known as it not working

Bloomberg @business

7 days ago

“The technology worked. The value didn’t arrive,” Bain concluded in the report. bloomberg.com/news/newslette…

36 137 680 487K 402

Heads up, there is no bubble Infra and hardware are not a bubble Don’t wait for anything to burst and work towards tangible goals with that expectation in mind

Ahmad @TheAhmadOsman

8 months ago

“i am waiting for the ai bubble to burst”

40 24 388 32K 13