Backscroll ranked feed — 2026-07-06 02:47Z

47fucb4r8curb4fc8f8r4bfic8r @47fucb4r8c69323 2026-07-05 20:15Z

I hate people, I am cynical, I am easily disgusted by others. But humanity is the only life form we know of that can find another species cute and be motivated by that to want to care for, provide, nurture and bring happiness to that creature. The cynic says this is unfair to https:// x.com/DanielleFong/s /DanielleFong/status/2073847647035965885 …

tomie @tomieinlove 2026-07-06 01:20Z

There’ll be a point where AI development briefly stalls as researchers, not comprehending what their checkpoints are saying, assume something must’ve gone wrong in the training run.

Jukan @ ICML @jukan05 2026-07-05 13:41Z

According to a new review of the ThinkBook 14 G9 by Notebookcheck, Chinese PC maker Lenovo has, for the first time, begun shipping globally configured laptops equipped with YMTC-made SSDs. $SNDK $MU

nico @nicochristie 2026-07-05 21:38Z

I’m no macro economist But there are a lot of 39 year olds grinding leetcode at Starbucks rn

Riley Goodside @goodside 2026-07-06 02:26Z

Work in progress, testing Fable 5: screensaver with infinite procedurally generated VHS found footage of someone lost in the Backrooms

Ethan Mollick @emollick 2026-07-06 02:27Z

Super weird thing about Fable is that it keeps learning about me even with memories and personalization turned off. The thinking traces are full of references to my preferences. Some of it is little notes it writes in various .md files, but maybe it is leaking from elsewhere.

Aseel Swaid @aseelswaid9 2026-07-05 22:50Z

Fishermen, Scarborough Beach, 1965

Architectural Art (A-A) @4AAAAart 2026-07-05 19:11Z

The Bascule lounge chair by Studio Œ for Vitra Photography: Benjamin Lund

Lisan al Gaib @scaling01 2026-07-05 23:34Z

the other thing im noticing while working on my research projects is how limited these models are GPT-5.5-xhigh is not even close to being an automated researcher you should also not ask these models for advice on how to design experiments. the models have 0 taste

Phillip Yan @PhillipYan2 2026-07-05 23:15Z

the protocol-as-universal-remote pattern is moving faster than i expected. curious whether the latency on editor control is tolerable for anything beyond asset placement or scene setup

Nathan Flurry @NathanFlurry 2026-07-05 21:46Z

throwback to my Waymo being shot at with fireworks, then quoted @Reuters international they tried to spin it in to FSD hit piece validating the assault of Waymo passengers as a form of debate

Andrew Curran @AndrewCurran_ 2026-07-02 20:54Z

On the heels of reports that META is exploring a move into compute-as-a-service like xAI, Mark Zuckerberg told an internal town hall that AI agent development over the last four months hasn’t accelerated 'in the way we expected'. The race continues to narrow.

Timothy Luong (Chongz) @chongz 2026-07-05 23:54Z

Certain sub-species of tech bros have evolved to identify more with work than family. I recommend immediately masquerading as an investor rather than an operator. By doing so you signal that you aren’t “in the arena,” diffusing their competitive instincts.

Susan Zhang @suchenzang 2026-07-03 13:52Z

books3 was absolutely necessary for llama to happen back in the day. you can even find my pathetic cleanup code for it here: https:// github.com/facebookresear ch/metaseq/tree/main/preprocessing/books3 … it was also noted to be a critical datasource for mmlu in the llama1 report, though by now we know mmlu is useless and high

Nikunj Kothari @nikunj 2026-07-05 22:54Z

Still waiting on that one founder who gates a VC chat until they have played with the product and come with at least two pieces of feedback.. Blows my mind that people go on Zoom and chat for 30 mins on a parroted deck for literally the entire day - where it could have been a

hope hopes hoping @hopes_revenge 2026-07-05 19:07Z

there should be a “born again” movement where big posters abandon their high-follower accounts and begin again as lowbies . walking amongst the scum and detritus

will depue @willdepue 2026-07-05 06:57Z

what sam is working on is the closest thing i’ve ever seen to what spacex must have felt like in the early days. considering joining him, it’s a truly ridiculous and incredible mission

Bin Shao @ShaoBin_phy 2026-07-06 01:14Z

Introducing a weekend project: Folding Everywhere. a pure-Rust, dependency-free re-implementation of ESMFold1/2 that predicts protein structure on the CPU with one executable, no Python, no PyTorch, no CUDA, and nothing to install!

broadglow @broadglow 2026-07-06 01:57Z

This isn't totally wrong but it also makes some assumptions. If you're clearing 1m+ at some bank/pe firm/biglaw job, you probably: - Get invited to destination weddings (flights, clothes, hotels, gifts) + bachelor/ette parties; each one is maybe a ~2-5k expense (suit/dress that

Kaifeng Lyu @ ICML @vfleaking 2026-07-06 02:04Z

When exploring methods for knowledge injection through synthetic data generation, we found that as the token budget scales up, these methods fail to outperform a very simple baseline: Repeatedly prompting an LLM to rewrite source documents using prompt templates sampled from a

Rohan Sawhney @rohansawhney1 2026-07-05 22:42Z

New WoSX demo: interactive geometric deformation with Walk on Spheres! Computes stochastic barycentric coordinates for detailed triangle meshes inside coarse cages, without tetrahedralizing the cage interior, then transfers interactive cage edits to the embedded geometry.

akira @realmcore_ 2026-07-05 23:46Z

4.8 is a pretty good orchestrator but frankly it doesn't really matter Orchestration decision making will be baked at the model layer within a few months

zan @xenoaesthetics 2026-07-05 20:47Z

Similarly, “what is ignorance” is more interesting than “what is intelligence”. Artificial Ignorance, universal deconstructors, etc. But is this mere pedantry, a rhetorical trick? Or is there any actual difference between the two approaches?

Lucas Beyer (bl16) @giffmana 2026-07-05 19:36Z

It's surprisingly bad still. I just had a session with my son, asking it to say the same sentence in various languages and it was 1) wrong like a third of the time, literally confusing languages (!) and 2) horrible accent in most languages.

max! @maxsloef 2026-07-05 23:43Z

base models are ontologies, not simulators

Aseel Swaid @aseelswaid9 2026-07-05 21:47Z

Florette, Paris, 1944

Jan Dubiński @jan_dubinski_ 2026-07-06 01:47Z

Presenting at @icmlconf ! Tue, Jul 7, 2–3:45 PM KST “Jailbreaking VLMs Through the Visual Modality” Frontier VLMs can be jailbroken by making them recover unsafe intent from visual context! At #ICML? Let’s talk about VLM safety. w/ @AharonAzulay , Zhuoyun Li, Atharv

Phillip Yan @PhillipYan2 2026-07-05 23:16Z

the mass ratio is wild but the politically relevant number is probably the energy density ratio, you're moving 300x the weight but also moving it to a furnace, not a factory that reuses it next cycle

Yifeng "Evan" Wang @ewind_dev 2026-07-06 02:17Z

I'm thrilled to share with everyone that, powered by Fable, I've finally created the ultimate cross-platform UI framework from my years of imagination—PocketJS! Porting QuickJS to the Sony PSP with just 32MB of memory, combined with standard SolidJS and Vue Vapor, and paired

Irrational Analysis @insane_analyst 2026-07-05 23:15Z

Two hyperscaler NPO programs and AMD's advanced optics program are all starved for lasers. Compare my InP model with SA model.

lusso @luusssso 2026-07-06 00:14Z

“Gherpe” table lamp by Superstudio for Poltronova (late 1960s)

Rhys @RhysSullivan 2026-07-05 19:05Z

this onboarding idea from @ExaAILabs is brilliant, get the user to complete your setup via offering credits i generally skip onboardings but this worked on me

Perly @_Shark_byte 2026-07-06 01:27Z

Summer in nyc means the invasion of the Waterloo interns

R.F. Kenmore @rfkenmore 2026-07-05 18:02Z

Todd-ler Boyfriend

Guilherme Penedo @gui_penedo 2026-07-05 22:32Z

Super cool blog! The impact of different intervention strategies (small nudge, rewind, letting it fail then correcting) and how policies start trying to correct already successful demonstrations is super interesting Is there an equivalent to LLM's EOS? Could help maybe?

Kendric Tonn @kendrictonn 2026-07-05 22:58Z

It's early; my personal prediction is that we'll get 90 people mentally stunted to hilarious degrees, 9 adapted to a new environment & trucking along vaguely as before, and 1 freak with freak parents who's tutored in the hills by MegaAristotle and that's when it gets real weird

BioTender @Junioryu136689 2026-07-05 20:34Z

Benchmarking AlphaFold2, RoseTTAFold2, ESMFold, OmegaFold & DMPfold2 on short peptides (10-49 aa) — a regime all these models struggle with. Key findings: accuracy scales with length, α-helical peptides predicted better than β-sheet, and intrinsically disordered regions remain

SightBringer @_The_Prophet__ 2026-07-05 22:24Z

Haaland is the purest specialist football has ever produced. He deleted everything from his game. The dribbling, the buildup, the flourish. All of it liquidated and reinvested into a single function: ending attacks. A goal per game at Salzburg. At Dortmund. At City. For

Shayan Talaei @talaei_shayan 2026-07-06 02:03Z

Suppose you're handed a fine-tuned LLM that secretly favors a certain entity. The bias goes completely undetected because it only surfaces on one specific unknown topic. So how do you catch a bias you can't search for? You amplify it. Introducing Distill to Detect (D2D), our

Malte Ubl @cramforce 2026-07-05 07:56Z

It's not widely understood that the implementation of the idea of a nation is roughly as old as the United States

Damian Barabonkov @damian_b 2026-07-05 23:15Z

Fable is a double-edged sword. From my experience this week, it is really good if you know the correct direction to go in. But if you don't, it will take you down a dead end, and all you'll do is spend thousands on tokens. Most enterprises will sadly venture down path two.

ArtButMakeItSports @ArtButSports 2026-07-05 22:46Z

Children's Games (detail), by Pieter Brueghel the Elder, 1560, by @kgmadeit1

ertdfgcvb 𓅰 𓅬 𓅭 𓅮 𓅯 @andreasgysin 2026-07-05 08:43Z

Link² (2025) Editions #4, #16 and #20, part of the @arabbankCH digital art collection. The “Link” series consists of mechanically generated animations created using long-exposure photography.

kache @yacineMTB 2026-07-05 23:46Z

Beat mujoco's g1 baseline by 10x. Trains in 3 minutes

Dan Gray @credistick 2026-07-05 17:19Z

If you apply the logic that private companies are 48–50% overvalued*, that puts Databricks at $111.5–110B. Seems about right, and a good comp to Snowflake. (* https:// nber.org/system/files/w orking_papers/w23895/w23895.pdf …)

Tongyi Lab @Ali_TongyiLab 2026-07-06 02:33Z

We love seeing the community push the boundaries of our Qwen-Image-Edit-2511!

Retro Gamer Art @RetroGamerArt 2026-07-05 21:31Z

Capcom Fighting Evolution (2004) Artist: Shinkiro 森気楼

GEOFF @geoffreywoo 2026-07-06 00:00Z

AI evals are cute until procurement asks the cursed question: who pays when the agent is right 97% of the time and the 3% failure happens inside a regulated process? that answer matters more than another leaderboard bar chart.

Architectural Art (A-A) @4AAAAart 2026-07-04 16:59Z

Stòffa Soho Flagship Store by Stòffa & Colin King New York, USA

Wiesław Šoltés @wieslawsoltes 2026-07-05 16:07Z

Desktop UI testing using Selenium? Yes, it works! I just brought full Selenium support to our custom Chrome DevTools Protocol (CDP) server for Avalonia (other frameworks to come)! You can now automate desktop apps using standard web testing tools (Selenium, Playwright,

Guohao Li @guohao_li 2026-07-05 19:35Z

more neolabs are emerging in China, founded by solid researchers. congrats @YongchaoC !

Sólionath @Anarseldain 2026-07-05 20:57Z

I’m working on a new unified theory of American reality called “Everyone Is 40 Now.” “I’m scared of new things” of course you are. You’re 40. “I just don’t understand these weird youngsters” yeah dude, you’re 40. “You’re not a badass knight, grow up.” Oh you old fart, you’re 40.

Alex Shaw @alexgshaw 2026-07-05 00:59Z

An under-appreciated aspect of Harbor is that results are auditable and reproducible so you can make trust-less claims about your agent or model. In the era of SWE, code was the source of truth so OSS built trust. In the era of agents, evals are the source of truth and

alth0u @alth0u 2026-07-05 08:02Z

the hardest part to achieving perpetual motion is hiding the engine

August-–—• @augustaghast 2026-07-05 21:01Z

The last snakehandling church in West Virginia

Cyrus @cyrusnewday 2026-07-05 22:01Z

hot take: the best agent users are going to look more like RTS players than prompt engineers scouting, task queues, map awareness, not over-microing, knowing when to expand if you can keep 6 half-dumb units productive at once you’re cracked

Diyi Zhu @communidiyi 2026-07-05 22:31Z

in addition to the cafe compute party we're hosting at ICML (4600+ RSVPs and counting ), we bought out cafe gabdeo inside COEX for 3 days where @cerebras and @OpenAI team members are meeting with #icml attendees let us buy you a coffee. we're here tuesday wednesday and

jianlin.su @Jianlin_S 2026-07-06 02:23Z

Revisiting Convergence Results in Convex Optimization (VII) https:// kexue.fm/archives/11804 - New identity quantitatively link weight averaging and learning rate decay - discuss why uniform averaging loses to sliding, and why Schedule-Free still can't fully escape scheduling

Alice Evans @_alice_evans 2026-07-05 18:55Z

Using GPS-linked wearable cameras and randomised street audits across 900 kilometres of roads in Greater Mumbai, taking 4000 street images of 23,000 visible individuals, They find that women only account for 16% of visible people in Mumbai and 15% in Navi Mumbai. BRILLIANTLY

Aseel Swaid @aseelswaid9 2026-07-05 21:53Z

A bronze head of President Saddam Hussein is paraded in the streets after US troops occupied the capital, Baghdad, Iraq, 2003