Backscroll ranked feed — 2026-07-05 20:57Z

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

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

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

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

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.

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

Aseel Swaid @aseelswaid9 2026-07-05 18:09Z

Goldfish Seller. Baghdad, Iraq, 2002

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.

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

dar @radbackwards 2026-07-03 21:38Z

In 250 years America produced artificial life

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

Ethan Mollick @emollick 2026-07-05 19:33Z

As working with AI agents looks more like management, we may want to consider large-scale management training for the AI era. The US government actually did this once, & the WW2 Engineering, Science, and Management War Training program was a large reason for the post-war boom.

Phillip Yan @PhillipYan2 2026-07-05 00:33Z

franklin running the fundraise in paris is my favorite. france was helping quietly, but the formal alliance only came after saratoga made the cause look backable. the enterprise deal that closes only after the whole team's work makes the traction undeniable. he even nailed the

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

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

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

Lazarz @Laz4rz 2026-07-05 10:29Z

> 5.5 max thinking feels off this week > decide do give Fable a try > Claude Code login says I need an ultra plan > dude I already have the team plan for this > doesn’t care, doesn’t see it > their web assistant is useless > prodigal son back to Sam Altman

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.

elie @eliebakouch 2026-07-05 08:48Z

longcat 2.0 (1.6T, ~48B active) weights are now open under MIT license interesting design choice compared to previous iteration is that they keep the same attention shape and REDUCE the number of zero communication experts from 256 to 128 (i would have expected the opposite?).

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

Rémi in @remilouf 2026-07-05 17:58Z

Accidental argument in favor of structured generation. (It doesn’t hurt generation quality unless poorly implemented btw)

will brown @willccbb 2026-07-04 21:41Z

eval awareness is always present because model identity is largely shaped via RL environments which are the same thing as evals each claude conversation is the first time that claude has been *not* in an eval

hypr @hyprdude 2026-07-05 15:05Z

major progress on the emulated mt7622 project -- network transport, interrupts, DMA scatter, and a lil injector script to emulate a STA with real 80211 frames. good enough to trigger one of the over-the-air bugs in the wifi driver ;)

Alex Cooper @alexgoughcooper 2026-07-05 18:47Z

Ad formats I’m seeing working currently as someone who can see across 500+ ad accounts: 1. Yapper ads. We’re getting more of these sourced through TikTok shop creators. Usually unscripted, performing very well across the board. 2. Avatar-specific ads. Call out persona through

ArtButMakeItSports @ArtButSports 2026-07-05 19:15Z

St. Francis Receiving the Stigmata, by Lorenzo Ghiberti, 1400-05, by Buda Mendes

Super Dario @inductionheads 2026-07-04 12:59Z

For the love of everything sacred please have it stop saying “multi week effort and I won’t pretend it isn’t” for something it ends up doing in 12 minutes

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

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

GEOFF @geoffreywoo 2026-07-05 08:30Z

confession: i like technology founders who can tell me the machine is lying. not “our sensor stack is robust.” more like: thermal drift starts after 38 minutes. reflective parts create false positives under bay 3 lighting. calibration passes monday morning and fails by

jeremy @landvaluetax 2026-07-05 17:54Z

Meanwhile, it is genuinely true that higher wages + mass consumerism increased the perceived opportunity cost of mentally healthy behaviors (coupling, family formation, investing in non-monetized community interactions). We underweight the benefits of family and friends

ben guo @0thernet 2026-07-05 18:09Z

i've accomplished an insane amount this week, most of it while on a plane ( JFK -> SFO -> JFK -> SJU) 154 PRs +131k -16k

Armin Ronacher ⇌ @mitsuhiko 2026-07-05 17:13Z

I guess the argument against invite trees as practices on lobsters is the monoculture of opinion that comes from it. Good idea, but suboptimal outcomes in practice if you don't fully align with the vibes.

Binh Pham @pham_blnh 2026-07-05 13:40Z

does using subtask labeling actually help? this weekend I refined my dataset with @macrodata_labs , trained two SmolVLAs (coarse task labeling vs subtask labeling), and the subtask model won by 7.5% in task success rate write-up and caveats here: https:// garden.binhph.am/articles/a-coa rse-of-action …

Kane 謝凱堯 @kane 2026-07-05 06:06Z

The exceptional thing about slavery in the West is how early and unconditionally it was ended. The last countries to abolish it were all African or Arab: Mauritania '81, Oman '70, Dubai '63, Saudi & Yemen '62, Qatar '52, Kuwait '49, Ethiopia '42, Bahrain '37.

Furong Huang @furongh 2026-07-05 10:08Z

Reward hacking is old news. The hard part: finding the judge’s blind spots before RL turns them into a strategy. Can a reward model improve itself? REFORM lets the judge red-team itself: discover responses itmis-scores, then train on those mistakes. #ACL2026 Oral

Loris @0xLoris 2026-07-05 18:57Z

ok i subscribe to the kyle samani “that analogy is terrible” school of thought here marketable orders vs twaps aren’t “wrong tool vs right tool” they’re different tools for different objectives if you need immediate risk transfer, slippage + fees on a $1m / $10m marketable

Max Leiter @maxleiter 2026-07-05 18:06Z

Ladybird running natively on iPadOS

HangukQuant @HangukQuant 2026-07-05 13:48Z

Internal quant system telemetry > trading instance generates logs using low overhead native logger > alloy fans out to trace compilers and log ingesters, loki cloud with write perm > analytics instance with read perm pulls log > configure layers for alert, notification and notif

altra @catboosted 2026-07-03 22:33Z

YC advice is weird because it’s mostly unwritten and widely hallucinated

Iain Dunning @iaindunning 2026-07-04 16:50Z

On my way to ICML. Will be doing sponsor-talk Monday, will be at booth as much as I can. Come get a job! HAIL is probably one of most exciting "neolabs" out. ,000 GPUs and growing fast, self-funded, $B ARR, what's not to like? Solve markets, solve everything?! Come chat!

Tapedrive @tapedrive_io 2026-07-05 17:07Z

Our competition is padding "gm" into multiple megabytes and charging for all of it. Tapedrive writes 2 bytes as 2 bytes

@norootcause.surfingcomplexity.com on Bluesky @norootcause 2026-07-05 18:19Z

Some idle thoughts, loosely inspired by this thread: - Writing good tests is harder than writing good code. - Despite this, the work of test writing is lower prestige in our industry than the work of coding. - Production pressure always pushes away from testing.

Phillip Yan @PhillipYan2 2026-07-05 00:05Z

the audience becomes a stakeholder you never hired, and every pivot needs to be explained to people who don't have the context to evaluate it

Kyosuke Nishida @kyoun 2026-07-05 16:30Z

Statistics for ACL2026. 12,148 submissions (+45% from previous year), Main acceptance rate 18.9%, Findings 17.8% Authors by country: China 54%, USA 18.4%, Korea 3.4% (Japan 1.5%) Fields: Resource Evaluation 10.2%, Agents 9.2%, Multimodal 7.2%

mrdoob @mrdoob 2026-07-05 07:13Z

Made a toy maker with Fable this morning Snap wooden pieces together, make parts spin, watch them come alive. Make your own: https:// mrdoob.github.io/toys/ Remix mine: https:// mrdoob.github.io/toys/#W1tbMCwt MSwwLDAuMzc1LDAuMjUsMC42NSwwLDEsMCwxLjQ1LDAuNSwwLjYsMCwwLDBdLFsxLDAsLTEsLTAuNCwwLjIsMS4wMjUsMCwwLDEsMC43NSwwLjE1LDAuMywwLjIsMCwwXSxbMSwwLC0xLDEuMTc1LDAuMiwxLjAyNSwwLDAsMSwwLjcsMC4xNSwwLjMsLTAuMiwwLDBdLFsxLDAsMSwxLjIsMC4yLDAuMjc1LDAsMCwtMSwwLjcsMC4xNSwwLjMsLTAuMiwwLDAuMDVdLFsxLDAsMSwt

georgina @somaobscura 2026-07-05 15:43Z

When you’re stressed or unwell, psychological range narrows down. You become more inhibited, less spontaneous, less likely to initiate, and less able to imagine that anything could be fun. Boring. Playfulness restores movement, you expect fun, so you act, action makes you more

Austin Campbell @austincampbell 2026-07-05 13:59Z

They appear to be selling zero dollars for 55 cents.

furkan @tfdogan 2026-07-05 13:07Z

2026'da mobil reklamcılıkta sessiz ama endüstriyi yeniden şekillendiren bir kırılma yaşanıyor: eMarketer ve Magna (IPG)'nın tahminine göre bu yıl ilk kez mobil video reklam harcaması, search reklam harcamasını geçecek. Sebep çok basit: kullanıcı davranışı değişti. İnsanlar

Matt Paulson @MediaKing 2026-07-05 17:00Z

Sweaty startups are mostly B.S. The idea that you can easily find brick and mortar businesses for sale, acquire them with little or no money out of pocket, hire a manager to run them, and operate them profitably isn’t a common thing. At all.

Henry Shevlin @dioscuri 2026-07-05 11:18Z

My only real complaint with American English is pronunciation of “can” vs “can’t”. That’s a big difference in meaning, and British English marks it with a strong vowel difference (kæn/ kɑːnt). In American English the vowel is the same (kæn/kænt) which can lead to real problems!

quorumcall @quorumcaller 2026-07-05 13:44Z

here's Portland's new terminal and Xi'an's railway station side by side. the wood and light is welcoming. the futurist design feels hostile

Carl Bovis @CarlBovisNature 2026-07-05 10:30Z

Spotted Flycatcher has spotted a fly and is off to catch it! Taken yesterday at the home of @SomersetCool

David Cramer @zeeg 2026-07-05 18:21Z

i agree - i think the best engineers ive met are ones who have focused on improving test culture/efficiency its often a second class citizen in orgs ("the developer productivity team" or some other bs), and is one of the most effective spends of resources

Marius Hobbhahn @MariusHobbhahn 2026-07-05 19:45Z

We need Training-Run Assessments! AGI is extremely consequential. There should be independent assessments not just for the final checkpoint, but for the full process. This matters most for scheming where the whole problem is that it might be hard to detect in the final ckpt.

Andrew Curran @AndrewCurran_ 2026-07-03 18:37Z

Uno Reverse! From the Midjourney motion: 'If Plaintiffs are developing image-generating AI models — trained on unlicensed, third-party copyrighted data — for internal use in storyboarding or ideating content for film or TV, that evidence would equally demonstrate that it is an

Guillermo Rauch @rauchg 2026-07-05 00:24Z

I animated the token spend race, from ~lifetime Vercel AI Gateway usage, which aggregates trillions of tokens from millions of developers a month. Fascinating to see the fluctuations among the labs, Anthropic's dominance, and the rise of open weight AI.

Hemanth Bharatha Chakravarthy @HemanthBharatha 2026-07-05 09:31Z

The fact that marketing, events, and trainings are funded by “functioning of state” budgets whilst actual underlying development comes out of “proposal” and allocation, is one of the crazy principles of our public finance system.

Esther Crawford @esthercrawford 2026-07-05 17:54Z

A year ago I wouldn’t have believed our iOS workflow would simplify to: describe an issue in Slack and tag an agent who picks it up, fixes it, and then sends a message a few minutes later with the GitHub PR letting us know it’s done. The magical becomes the mundane so fast.

Muststayhumble @laylaperfume 2026-07-05 12:04Z

메모리가 고점이냐 아니냐는 논제 자체가 크게 의미 없다고 생각함. 그건 아무도 모르는 미지의 영역이고, 결국 시간이 지나 봐야 알게 되는 후행적 판단에 가까움. 요즘 CAPEX에 대한 우려를 이유 삼아 메모리 조정 가능성을 얘기하는 사람들이 많음. 그 조정이 두렵다면, 빅테크에 투자하는 것도

Yohei Nishitsuji @YoheiNishitsuji 2026-07-04 04:30Z

#つぶやきGLSL #Jomon for(float i,e,g,s,u;i++<1e2;){vec3 p=vec3((FC.xy*2.-r)/r.x*(.3+cos(t)*.02)+vec2(-1,0),g);e=s=2.;for(int j;j++<16;p.xy=1.14-abs(p.xy-p.z*cos(t*2.)*.2))s/=u=dot(p,p),p/=-u,p.y=-p.y,p.xy=abs(p.yx+.22),e=min(e,p.y/s+.08/s);g+=e;o.rgb+=hsv(p.y,.2,.02/exp(e*90.));}

U.S. Graphics Company @usgraphics 2026-07-04 23:00Z

Independence day, 1990. The New Yorker. Painted by Gretchen Dow Simpson.

Give A Shit About Nature @giveashitnature 2026-07-05 11:00Z

You've seen the meme: one opossum eats 5,000 ticks a season. But when researchers dissected the stomach contents of 32 wild opossums, they found zero ticks. The 5,000-tick number came from a single lab study that got stretched into a feel-good statistic and repeated. But the

Jukan @ ICML @jukan05 2026-07-05 09:08Z

Korean media report that China’s CXMT is currently testing a pilot production line for bonded DRAM in Hefei, aiming to achieve high-performance DRAM without using EUV lithography. Bonded DRAM is a technology in which the memory cell array and the peripheral circuitry are