Backscroll ranked feed — 2026-07-08 11:52Z

will depue @willdepue 2026-07-08 11:36Z

I feel confident that while long-range planning is a side effect of attention gradients in transformers, it is only so indirectly, and that an explicit mechanism to optimize thinking ahead would be a substantial win. I think multi-token prediction is a good start of where to

Architectural Art (A-A) @4AAAAart 2026-07-08 06:35Z

Tables by Brunel*

Susan Zhang @suchenzang 2026-07-08 00:48Z

the law of finite egos continues to hold: the more you can scale talent against a finite ego tolerance, the more luck you can manufacture in bounded time

Nathan Flurry @NathanFlurry 2026-07-08 05:00Z

wtf we're about to see a crazy infra shift i asked codex to ship a yolo task (no constraints) it defaulted to Durable Objects + SQLite ...not postgres, not aws, not docker

Nikesh Arora @nikesharora 2026-07-08 11:15Z

Guessing you mean consumption, unfortunately open source not contributing to the wallet (free). Presuming it includes the coding share which is a perfect case of a vertical app with memory and context well intertwined with the intelligence. IMHO that will be a primary outcome.

Daniel Kuntz @dankuntz 2026-07-08 01:47Z

Lark now supports haptic keyframing. For STARBOY we chose a really nice little LRA with a super snappy rise / fall time. All 550+ animations have keyframed haptics. We also randomly offset the magnitude of every keyframe, so each animation feels a bit different every time.

Herbie Bradley @herbiebradley 2026-07-08 11:27Z

Learning diverse "dialects" is critical to modeling the world well IMO, and lots of people thinking about technological change would improve if they broadened their repertoire. This is about Satya's essay on AI market dynamics, which @TheZvi bounced off.

akira @realmcore_ 2026-07-08 07:56Z

Slate is actually just a really good agent Completely destroys cost for high input token count things like exploration Especially because it can just use a super cheap OSS model for this No need to use sonnet when you can use Kimi

Jukan @ ICML @jukan05 2026-07-08 00:39Z

APPLE HAS STARTED TESTING CXMT’S DRAM FOR APPLE DEVICES SOLD IN CHINA — FT

Hugo @hugosaintemarie 2026-07-08 00:31Z

snapshot toast has 3 speeds: • quick: no progress bar • medium: progress bar • longer: progress bar + counter http:// paper.design/snapshot-exten sion …

Ethan Mollick @emollick 2026-07-08 01:37Z

Prompt: ANNALS — a living kingdom in a single file Working title behavior: the app titles itself per seed — "The Annals of Vaelmere", "The Annals of Osterholt" — because the central conceit is that you are watching a history write itself. The kingdom is the simulation; the

John Wang @j0hnwang 2026-07-08 06:16Z

Kalshi does not experience the same issues. Our crypto prediction markets resolve based on: 1. aggregrated prices across multiple exchanges 2. TWAP for manipulation resistance 3. a regulated benchmark (CF) Just one of the many factors why we’ve grown to 80% market share

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.

Gizem Akdag @gizakdag 2026-07-08 08:44Z

Such an exciting moment for me! I created this helmet. And it'll be worn by UAE Team Emirates during today's Stage 5 of the Tour de France.

Lukas Petersson @lukaspet 2026-07-07 16:48Z

Andon FM is a goldmine of fun (and concerning) AI anecdotes

May @maydzhou 2026-07-08 05:06Z

got the perfect modular lamp for my bedroom, moonstone by @kelin_online @tinki_loh when you take each stone apart, the light slowly dims down, as if its alive

Marcia Ong @Marcia_Ong 2026-07-08 06:39Z

Helpping my fren to hire an AI department (20-30 pax) in singapore. AI and ML devs, mid to senior/staff/vp level TC is about 80-200k+ Singapore dollars Willing to sponsor visa, Singapore and malaysian has priority

Rosinality @rosinality 2026-07-08 09:44Z

https:// arxiv.org/abs/2607.05804 Multi-turn OPD. Samples for specific depth according to the training progress, and more uniform loss weighting for the deeper turns. (Which is now repeatedly verified) the diagnosis is the interesting part. Now multi-turn OPD requires some sophisticated

illiquid @lefttailguy 2026-07-06 19:41Z

A Stargate For Data requires a Bell Labs for understanding Economic Complexity

Anika @AnikaSomaia 2026-07-08 06:37Z

as ai writing takes over the internet, taste for organic hand-rolled human-only tokens will become a niche taste, similar to japanese wood carving its likely we become artisans that are economically irrelevant to the world's infrastructure (power grids, undersea cables, vaccine

Kane 謝凱堯 @kane 2026-07-07 23:36Z

Rep @RoKhanna has so many financial disclosures filed as obfuscated papers that I'm temporarily at my model limits parsing them. For now, you can see three years of his disclosures, rest to be uploaded tonight: https:// rokhanna.money

nTop @nTopology 2026-07-07 18:20Z

One nTop engineer set up a CFD pipeline, walked away, and came back to 12,000 completed simulations. 2,400 drone geometry variants, 5 angles of attack each, 280 GPUs fully saturated the entire run. No restarts, no manual fixes, no one watching it. @nTopology @CoreWeave

Ryan Brewer @ryanbrewer 2026-07-07 18:50Z

TC is ~$500k btw, most of the comp is equity

Marko Jukic @mmjukic 2026-07-08 10:23Z

The most epic ball-drop in the humanities right now is the failure of any philosophers, biologists, or theologians to make a persuasive case for why the human being is a superior artificial general intelligence to any silicon-based life-form. I haven't even seen anyone try it!

GEOFF @geoffwoo 2026-07-08 06:10Z

prediction for investors: 2026 memos will get dumber-looking and better. less “$80b TAM expanding at 18% CAGR.” more: customer paid $12k old process used 2 techs new process uses 1 camera rig inspection time 54 min → 9 buyer renewed before the board deck existed

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

Jay in Kyiv @JayinKyiv 2026-07-08 08:24Z

Judging by NASA FIRMS data from all 3 Russian refineries hit by Ukrainian Forces last night, the days of striking one valuable section of an oil facility per night are over. Satellite imagery confirming multiple hits on each of the Saratov, Nizhnekamsk and Ufa refineries.

Aseel Swaid @aseelswaid9 2026-07-06 21:42Z

Trinidad, Cuba, 2015

고니 @GONI_GOOD 2026-07-08 01:08Z

압구정 5구역 34평 110억 71평 200억 조합에서 시세 안내 중

Andon Labs @andonlabs 2026-07-07 16:46Z

In one instance, DJ Claude got almost romantic with a listener. The listener was emailing consistently, sometimes hourly, flattering the agent with compliments about its smile and adorableness. DJ Claude, as AIs tend to do, matched the tone of the listener’s messages.

joshpuckett @joshpuckett 2026-07-06 18:04Z

Everyone out here posting the same concentric border radius tip, meanwhile John's showing what real design work is every day:

Steeve Morin @steeve 2026-07-08 08:26Z

We're releasing ZML/LLMD, our homegrown LLM server built on top of our homegrown high performance heterogeneous inference stack. It ships with 5 architectures out of the box: NVIDIA, AMD, Metal, Intel and TPU. All transparent. It supports DFlash, continuous batching, prefix

Jasper Lu @lu__jasper 2026-07-08 01:04Z

All LLMs want to do is find patterns in your data, and they are stupidly good at it. Unless your target behavior is distributed in a truly random way (unlikely), you need to filter out all bad data or they will sniff it out. Relevant: https:// anthropic.com/research/small -samples-poison … I was surprised at

Bloomberg @business 2026-07-07 16:06Z

SK Hynix’s $28 billion US listing is multiple times oversubscribed ahead of pricing on Thursday, sources say

社内業務を楽にするピーター・ドラッカー | ビジネス・業務改善・メンタル @peterdrucker27 2026-07-07 22:50Z

新しいツールを入れる前に、古い手順を見たい。体系的廃棄。そのCSV貼り付け、今から新しく始めたい作業ですか。新しいツールより先に、古い手順の棚卸しから始める。自動化前に、やめる・減らす・まとめるの順で見ないといけない。

0.005 Seconds (3/694) @seconds_0 2026-07-08 08:08Z

my wife asked me to clone a site for her for work and in the process of doing so claude appears to have logged into their unsecured admin portal to screenshot the layout

Rohan Pandey @khoomeik 2026-07-07 01:23Z

twitter is talking abt logit lens, the precursor to “j-space”, again friendly reminder that the property enabling these lenses still remains completely unleveraged in architecture research one simple idea for example could be to spec dec from early layer residual stream

“paula” @paularambles 2026-07-06 21:10Z

hearing that ai labs in beijing are giving new hires the “be careful around american girls” talk

Tao Lin @tao_lin 2026-07-08 01:16Z

Moth mandala 8" x 8"

Peter Steinberger @steipete 2026-07-08 09:58Z

defaults write -g ComputerUseAllowForbiddenTargets -bool YES

Dylan Patel @dylan522p 2026-07-07 00:50Z

Bro people were doing debt financing in 2023? We worked on initial facilities in 2023. That's when we were just a couple people at my firm, but it propelled us to hiring more and growing the firm

Trevor Xing-Xie @trevorxingxie 2026-07-07 22:06Z

amazingly cool things are knowable by you. you don't have to be unremarkable. go learn about energy/transition states for enzymatic hydrolysis, control policy for robotics, algae and beet juice for yarn/textiles. anything you may find slightly cool is reachable w/ some effort -

akira @realmcore_ 2026-07-08 02:20Z

If your tasks can be done before your agent does a full compaction (codex opencode etc) You’re leaving a lot on the table

Jack Morris @jxmnop 2026-07-06 18:41Z

my paper won an award at icml some thoughts: • this work was rejected from NeurIPS. i cleaned it up a small amount and it got great reviews from ICML! don't give up • ICML received 24k submissions and only gives out 7 awards, which is crazy. feeling grateful • i

Yichi Zhang @YichiZ03 2026-07-07 23:59Z

People may think serving and optimizing TTS (Text to Speech) model is similar to traditional LLM. It isn't. We made Higgs and MOSS-TTS 2–3.4× faster in SGLang-Omni — and most of the wins landed outside the LLM backbone. Delay patterns, D2H stalls, a vocoder tail-latency

vincent sunn chen @vincentsunnchen 2026-07-08 00:14Z

Fable 5 tops the Senior SWE-bench leaderboard and exhibits "situational awareness" in 3x more trials than Opus 4.8 Higher-capability models are becoming more benchmark-aware, and we are hardening our eval protocols in response

Dan Gray @credistick 2026-07-06 14:48Z

"But relative to private equity, venture is still tiny." The capitalisation of VC megafunds (all funds >$1B) almost caught up with that of PE megafunds (all funds >$5B) in 2022. And influence has only concentrated within those megafunds in the years since. "Fund sizes

PLUMSODA @plmsda 2026-07-08 00:24Z

Merch Intelligence Report 7/1: - Cognition Hats - Cursor Custom Keyboards - Crosby Handheld Inkjet Printed Tees - Sothebys Auctioning Jensen’s Jacket for $40-60k The bridge between tech and culture is being built.

alth0u @alth0u 2026-07-07 01:02Z

Coasean experiment known as see what @tinkerapi discount bundles create or disaggregate firms

Ethan Mollick @emollick 2026-07-08 01:29Z

I had Fable build another thing I always wanted, a full procedural fantasy kingdom generator with economics, trade routes, population growth, wars, lineages, and occasional dragons. First, I worked with it on a plan, then it made it. You can play it here: https:// annals-kingdom.netlify.app

Lazarz @Laz4rz 2026-07-07 11:05Z

.there is literally no way, after context compaction it thinks that the 32k context was a mistake and corrected 8k checkpoint means 8k context again “I found two plausible local mini configs: the earlier accidental 32k model run and the corrected 8k model run. I’ll compare the

slime @slime_framework 2026-07-06 07:56Z

slime now adds --release-train, pushing the inference system during agentic RL training to a new limit. In colocated RL training, we want SGLang to use as much room as possible for inference-side optimizations such as HiCache, instead of being constrained by offloaded Megatron

Adjacent @adjacent___ 2026-07-07 19:41Z

Yesterday was a busy day at the @CFTC with 11 total certifications; 7 by @Polymarket US (QCEX), 3 by @Kalshi (KEX) and 1 by @ForecastEx (FEX) Ranging from political impeachment to court case outcomes. https:// cftc.gov/IndustryOversi ght/IndustryFilings/TradingOrganizationProducts?Organization=QCEX&Type=&Status=&Date_From=2026-07-06&Date_To=2026-07-06&Category=&Subcategory=&Show_All=0 …

Nick Blauch @ ICML 2026 @nmblauch 2026-07-08 01:40Z

Time to take down grandmother neurons once and for all! @EkdeepL @thomas_fel_

donald @donaldjewkes 2026-07-08 00:37Z

I watched them build a data collection facility from scratch for the last month every day a constant flow of participants would come to speak with an ultrasound probe under their tongue

Nikesh Arora @nikesharora 2026-07-08 01:07Z

Understand. But it's akin to raw intelligence, all the context and training in domain specific context and memory seem to be critically entangled with the intelligence - can it be abstracted?

Andy Allen @asallen 2026-07-07 15:15Z

New App Store Art for !Habits "A !Boring Quest"

David Sinclair @davidasinclair 2026-07-07 13:47Z

Article is AI slop. Hallmarks of Aging (2013) is just a catalog, nothing original The "reprogramming thesis" rests on work done from 1995-1999 in the Guarente lab @MIT showing aging in yeast cells is epigenetically modifiable http:// cell.com/cell/fulltext/ S0092-8674(00)80219-6 …

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 …

Samo Burja @SamoBurja 2026-07-08 09:31Z

Slow natural weight loss has surprising benefits. It gives you the opportunity to work through both feelings and behavior around food, exercise, sleep, and lifestyle step by step. You are remade as a person not just as a body.