Backscroll ranked feed — 2026-06-25 09:30Z

Ethan Mollick @emollick 2026-06-25 02:58Z

The capability overhang from the models we have today is big enough that large-scale change to work and society over the next 5+ years is now inevitable even if AI development stops. (And there is no real sign that AI development is slowing down, it appears to be accelerating)

“paula” @paularambles 2026-06-24 23:19Z

whenever LLMs are kind of dumb at anything other than writing code it’s like wow it turns out you can be kind of stupid and still write good code

roon @tszzl 2026-06-25 01:22Z

‘ensure that AGI benefits all of humanity’ turned out to be a quaint, bearish statement. whatever AGI was, it was surely in the past or at most present, and its definition overly concerned with economics. “ensure that mankind can coexist with superintelligent machine ecology”

Tsung Xu @tsungxu 2026-06-25 03:54Z

Restrained ground testing begun. The speed at which we can identify and fix issues with both the prototype and our processes is just unreal. At least 10x if not 100x faster than how fast I was fixing issues on my own VTOL builds back in pre-agentic dev days.

cursed_connectors @cursedconnector 2026-06-25 06:27Z

So the connector that this guy is claiming isn't a standard plug in the US, absolutely is, as one of the IEC 60320 C15/C16 receptacles, which even has UL certification available under UL 498... or something. https:// en.wikipedia.org/wiki/IEC_60320 #:~:text=C15-,C16,-M …

Jukan @jukan05 2026-06-24 23:16Z

I find it hard to understand why some people are interpreting Micron’s latest LTA signing as the end of the memory price upcycle. Signing an LTA ultimately means that the amount of supply in the market that has not yet been locked up is shrinking. In other words, the later a

Matt Slotnick @matt_slotnick 2026-06-25 02:06Z

it's a seed round when you used to work at anthropic

Alexander Doria @Dorialexander 2026-06-25 08:25Z

Unfortunately our core issue is much worse than over-regulation: generalized grift. Policy people are daily joining mid companies they helped propel/entrench and it’s totally normalized.

Liang Zheng @LiangZheng_06 2026-06-25 02:24Z

DiffusionBench supports various auto-encoders, such as lots of RAEs, REPA-E family, VAVAE, and traditional VAEs. it also supports REPA, iREPA, MeanFlow, REG, internal guidance, DDT architecture, etc. using a unified data loader. supports GMuon.

Bhavya Bahl @bhavya_bahl 2026-06-25 07:09Z

Just going through the list of problems is a great way to become familiar with common distributed paradigms in LLMs

Stanford AI Lab @StanfordAILab 2026-06-25 07:51Z

Most open agentic datasets target one benchmark. In compute-controlled comparisons, OpenThoughts-Agent-v2 leads at every training set size, and generalizes across seven agentic benchmarks. Check it out!

Julia Love @byJuliaLove 2026-06-25 01:57Z

More on AI talent wars: Two more senior researchers are leaving Google for Anthropic, and Google shifted compute shortly before Noam Shazeer announced his move to OpenAI. Details from me, @nmasc_ and @rachelmetz , with help from @daveyalba and @mhbergen :

Matt Rickard @mattrickard 2026-06-25 02:34Z

Doing most of my code review locally with /review now. PR review bots feel wrong Have to pull the changes down and run them through the harness anyways. Suggestion is pre-CI. Too much noise, not enough signal, GitHub's fault too Problem is all the context gets lost for others

altra @catboosted 2026-06-25 06:45Z

The people raising tons of money right very well understand there’s an enormous capabilities gap between now and 1.5-2 years from now Sitting on a horde of cash and chilling is actually a winning move. Skate to where the puck is going

Chris Painter @ChrisPainterYup 2026-06-25 04:07Z

Jeopardy category that's like "Lesswrong post or San Francisco billboard"

Enze Xie @xieenze_jr 2026-06-25 02:31Z

Sol Video Inference Engine is here! An agent-native, training-free full-stack accelerator for video diffusion. It auto-tunes cache + sparse attn + token pruning + quant + kernel fusion for any model/hardware/config. >2× end-to-end speedup on 64B Cosmos3-Super, 22B LTX-2.3

Ag2S @Ag2_S1 2026-06-25 08:02Z

absolutely wrong, less character doesn’t means less tokens, that’s depends on tokenizer

stefan @wasserstein_rao 2026-06-25 04:20Z

Cel Shaded Assembly Viewer

Kenny Workman @kenbwork 2026-06-25 03:26Z

Human interpretation of agentic biology evals is difficult, especially for long horizon tasks. Real science admits multiple, branching analysis paths, including solutions not anticipated by benchmark authors. We've been developing new ways to understand complex trajectories by

Tsung Xu @tsungxu 2026-06-25 02:34Z

Our development process will save hundreds of thousands of dollars in proprietary but manual aerospace tools, said an engineer who has worked at one of these companies. Huge and widening variance in what toolchains companies are using right now

Neel Somani @neelsomani 2026-06-25 00:47Z

Releasing AutomataBench v1: Are models able to reason about dynamic systems under partial information? Real-world systems don't always give you all the information. AutomataBench measures exactly this - and the best frontier model reaches only 45%. http:// automatabench.com

Cristian Garcia @cgarciae88 2026-06-25 04:43Z

Refs allow expressing owned memory and mutation in JAX. They're both ergonomic and allow for optimization opportunities by constraining XLA.

haris @xPolarrr 2026-06-25 01:13Z

lfg, so bullish on the team at runpod, so happy to have been a part of the journey

(((ل()(ل() 'yoav)))) @yoavgo 2026-06-24 21:35Z

our language, ACDL (agent context description language) makes it easy to describe the structural semantic composition of an agentic prompt as it evolves over time. for example, here is ReAct context with tools at the end and history that includes reasoning traces:

Hyperliquid @HyperliquidX 2026-06-25 03:53Z

Portfolio margin on Hyperliquid has been live for six months and is now in beta with increased limits. Users with account value <$25M can use BTC and HYPE as collateral to trade perps, spot, and outcome markets with greater capital efficiency. See the Docs for borrow and supply

Jack Bookey @JackIsLearning 2026-06-25 02:35Z

The bin packing problem is why we will win The goal is traditionally to nest a set of items into the smallest amount of space. In our case it's cutting orders from our inventory of aluminum plate. It's an np-hard (computationally annoying) problem and at scale you can't brute

ParametricArchitecture @parametricarch 2026-06-24 21:14Z

Baobab Waterfall by Ahmad Eghtesad

Simon Willison @simonw 2026-06-24 18:53Z

Claude Code for web just started saying "GitHub is blocked by egress policy", which is a big problem for me because most of my prompts there start with things like "clone simonw/sqlite-utils to /tmp to see the docs in /tmp/sqlite-utils/docs" @ClaudeDevs @claudeai known issue?

gfodor.id @gfodor 2026-06-25 01:48Z

Alignment by default universe imo is the only real explanation for the fact the universe isn’t already computronium

Lewis Fausett @lewiswf 2026-06-24 15:41Z

This is why you are poor. $10m / year brand w/ 65% Gross margin @ 2.0 MER Cogs - $3.5m Ads - $5m Klaviyo + Shopify + SMS - $174k Misc Opex - $326k ($27k / month) To have $1m ACCRUAL net income. Which mean's you're likely not paying yourself. . .

Robert Johns @RobertJohns 2026-06-24 21:24Z

Just got off a call with someone who wants to bring the DC ecosystem to El Segundo — DARPA, Office of Strategic Capital, DoE, and Delta operators transitioning out and looking for their next mission. Would that be useful?

Ted Nyman @tnm 2026-06-25 05:54Z

the fact git has lasted this long (20+ years) is bonkers. statistically, something you wouldn’t expect to see in your lifetime. the messi of software. a testament to good engineering practices. an extremely careful set of maintainers & most of all: the C programming language.

Lucas Beyer (bl16) @giffmana 2026-06-24 22:17Z

It surprised me that the platonic representation paper got so much discussion, but this paper didn't. It's also one of the most severe academic burns I've seen in a while!

Dramz2(람지)/fevercell @nurik2ami 2026-06-24 10:43Z

Hatsune Miku Racing Queen 2020 >> 2026 I drew some fan art of Racing Miku! Hatsune Miku Racing Queen Fan Art #初音ミク #HatsuneMiku #レーシングミク #RacingMiku #ミクイラスト #MikuFanArt

Alex Nichol @unixpickle 2026-06-25 04:02Z

The salt compartment of the dishwasher was stuck shut. Three hours later, I opened it with a 3D printed, customized wrench.

Theo Jaffee @theojaffee 2026-06-24 15:59Z

Incredibly, San Francisco's local politics are now distinctly less radical/socialist than New York City's

Lex Tang @lexrus 2026-06-25 04:25Z

I just vibe-coded a water ripple shader for Ghostty with GLM-5.2. The GitHub link is available in the reply below.

LindyMan @PaulSkallas 2026-06-24 20:05Z

The weirdest part of reading about the british navy in the 18th century was that so many sailors could not swim, and swimming was not treated as a required naval skill. If their boat got damaged and they fell into the sea, rescue was unlikely and swimming merely delayed

Zephyr @zephyr_z9 2026-06-25 08:30Z

The only caveat here is that they won't use HCB till 2028

Justin Skycak @justinskycak 2026-06-24 22:38Z

Imagine a gym where everyone lifts according to birth year. That's basically how we organize education.

Hanming Fang @HanmingF 2026-06-25 00:39Z

High-speed rail may have helped fuel China’s EV boom by easing range anxiety and spurring charging rollout while amplifying its impact. HSR may account for up to one-third of China’s EV market-share growth, highlighting complementarities between intercity transit and EV adoption.

Hadley Harris @Hadley 2026-06-24 12:32Z

$3M ARR, can't raise an A, watching others raise $15M seed off a launch video

Fernando @zetalyrae 2026-06-25 06:59Z

How it feels to replace 160 lines of incomprehensible Claude slop CSS with 40 lines of human tasteslop.

Zachary Lipton @zacharylipton 2026-06-25 04:49Z

MXNet undead, resurrected by coding agents. Besides being a fun walk down memory lane, this is cool case study on the behavior & autonomy of coding agents & their capacity to animate legacy code (by @smolix ) https:// alex.smola.org/posts/46-mxnet -zombie/ …

gavin leech (Non-Reasoning) @gleech 2026-06-24 19:40Z

reviewing an obviously AI-generated paper (tone-deaf neologisms, 30 tables, 400 experimental conditions, vast fully-commented code). But Neurips has no rule against this, so I'm stuck with this burning sensation, wasting my bloody time. Then I see the hidden prompt injection...

Jenia Jitsev @JJitsev 2026-06-24 23:09Z

There is a lot of room to push open-source models. Eg using Qwen3-32B base with our composed data, we score on Terminal Bench 2.0 25.60% using only 100k samples and no RL, SFT only, outperforming Nemotron-3-Nano and coming close to Nemotron-3-Super that uses way more compute.

3X Long Labubu @labubu_trader 2026-06-24 21:00Z

Thanks to the Chinese government, $MU’s ER was stable. Chinese HFs used to be the best “sell the news” players to shake out retails and amplify volatility. Now they have to hold and “can’t sell high buy low,” as selling would mean losing the leverage to regain exposure.

th0rgal @Th0rgal_ 2026-06-24 07:59Z

Just reverse engineered my Oura Ring 5 so I can control my computer like a wizzard. @ouraring please send my love to whoever buried a feature to stream live accelerometer data

derek guy @dieworkwear 2026-06-25 00:36Z

If you look at photos of how actual American students dressed in the 20th century, you will find a much more eccentric and diverse expression of classic American collegiate style than what's popularly presented through Ralph Lauren ads (which are then recycled through Twitter).

San Francisco Chronicle @sfchronicle 2026-06-25 00:20Z

The San Francisco Civil Grand Jury has found the city’s homelessness department has failed to adequately monitor nonprofits that receive $500 million annually.

Tathagat @tathagat_ 2026-06-24 19:46Z

Built a better RAG with Proposition Indexing. Instead of embedding raw chunks, I embed key facts and retrieve full context later. 80% → 90% hit rate 895ms → 814ms latency 100% local stack. GitHub: https:// github.com/tathagat-git/p roposition-rag … #RAG #LLM #AI @rishabh10x @kirat_tw

Peter Sterne @petersterne 2026-06-24 20:44Z

I don't think young progressives even conceive of the idea that members of Congress are supposed to "bring money back" to their districts. They think in terms of national policy, like providing universal healthcare or childcare, not winning funds for specific pet projects.

Kai Cabero @kai_cabero 2026-06-24 09:00Z

I know a guy who pretends to be a venture capitalist to get his competitors to voluntarily hand over their entire client list, revenue numbers, and churn rate... He is not a VC. He has never invested in anything. He drives a 2019 Civic But he registered an LLC called "[City

Sheldrick Wildlife Trust @SheldrickTrust 2026-06-24 23:06Z

When an ex-orphan brings her new calf home, do our Keepers get to greet the baby? Sometimes – and it's entirely up to the mother. Some, like Nasalot here, walk straight up to a familiar Keeper and let them meet the calf. Keeper Benjamin had been a part of Nasalot’s journey

Kyle Kabasares @kylekabasares 2026-06-25 01:25Z

I may have gotten @OpenAI ’s ChatGPT-5.5 Pro to produce a candidate counterexample to an open problem from Don Knuth's The Art of Computer Programming. Exercise 210 in Volume 4, Fascicle 8A ( https:// www-cs-faculty.stanford.edu/~knuth/fasc8a. pdf …, page 55) asks a question about generating-function denominators

TBPN @tbpn 2026-06-24 19:30Z

President of @Blackstone Jon Gray says LLM spend among the company's portcos was up 15-fold in Q1 of this year over last year. At the same time, the pace of implementation is frustratingly slow, so it still feels like we're in the very early days of the AI cycle: "It's sort of

Lisan al Gaib @scaling01 2026-06-24 19:40Z

that's like perfectly in line with what I have been saying GLM-5.2 is as strong as Opus 4.5 and GPT-5.2 implying a 7 month lag

Sidhant Sibal @sidhant 2026-06-24 10:41Z

Passport is a travel document, not document of citizenship, say MEA officials

LanceDB @lancedb 2026-06-24 15:30Z

Updating a label on a row that contains a 50MB video shouldn't require Spark to read that video. Lance Blob V2 in Lance Spark keeps blob columns as descriptors through the plan. Bytes materialize only on write. A blob column stays a column. The SQL stays the same. What changes

Architizer @Architizer 2026-06-24 15:00Z

A 1915 log cabin sits at the heart of this home on the shores of Lake Wenatchee. Rather than demolish the aging structure, the architects preserved it in place and built a contemporary house around it. Now enclosed within a glass-lined shell, the original cabin functions as the