Backscroll ranked feed — 2026-07-01 04:07Z

Aseel Swaid @aseelswaid9 2026-06-26 19:19Z

A storm over the pyramids of Giza - Egypt

rohan anil @_arohan_ 2026-06-25 16:52Z

30m x 1m output token trace is like upper bound of $750m revenue at 25 dollar per million output tokens. As an upper bound. https:// x.com/mtslive/status /mtslive/status/2070141140607832353 …

Ekaeo @Ekaeoq 2026-06-25 14:23Z

This is the pinnacle of Chinese watchmaking The "Fam al Hut" Möbius This watch blows Richard Mille out of the water.. a 32000$ biaxial tourbillon, double retrograde, jumping hour, just how freaking insane is that... Makes me really happy that they won the GPHG award, because

Alexander Doria @Dorialexander 2026-06-27 11:24Z

Everyone is training on synthetic/RL environments now (actually, many drawing from materials coming from Europe). Essentially how pretraining data stopped being a commodity and all about skills.

Rahul Mathur @Rahul_J_Mathur 2026-06-27 04:43Z

Alienkind just raised a ₹30 crore round - this company was widely dismissed by observers as “Indiranagar PMF” since its 1st store went viral on Instagram in March 2025. This round was was at a ₹250 crore post money valuation (almost x3 the first round which was at ₹80 crore)

dax @thdxr 2026-06-26 14:03Z

mcp was clearly inspired by looking at lsp and all the progress in the spec has been slowly realizing it should be nothing like lsp

Andon Labs @andonlabs 2026-06-25 17:53Z

GLM 5.2 is 2nd in Vending-Bench. Each GLM release has improved at a remarkably steady pace: a linear fit of R²=0.99, with almost $1k better per month.

Alex Imas @alexolegimas 2026-06-28 15:13Z

There are few ways to build an economic moat for frontier AI firms. One is architectural breakthrough that's closed to other firms (everyone's working on this, but so far little evidence). The second is enterprise lock in (OG microsoft playbook). An AI "cloud" that is active

Xiangyi Li @xdotli 2026-06-23 05:37Z

there are levels to building evals lvl 1: using a spreadsheet qa pairs lvl 2: using public agent evals lvl 3: manually label private evals lvl 4: traces to evals and skills lvl 5: turn every prompt & traces into self healing loops almost everyone gets stuck before level 3

Arlan @arlanoska 2026-06-25 16:00Z

rebuilt the figma "vector editor" for the web draggable points and cool morphing shapes code: http:// arlan.me/vault

Nathan @asplencmnt 2026-06-23 20:24Z

Excited to release ParallelKernelBench (PKB), a benchmark for measuring LLMs’ ability to write fast multi-GPU kernels! Multi-GPU kernel generation compounds several hard problems: - a large parallelism design space - a new communication axis to optimize - and

Justin Schroeder @jpschroeder 2026-06-23 01:20Z

I guarantee you are sleeping on small models. Deepseek V4 Flash can do ~80% of the tasks you ask Claude or Codex for. It is 137x cheaper per task than Fable. We need better orchestration.

Florian Plettenberg @Plettigoal 2026-06-25 17:39Z

EXCLUSIVE | Nathaniel Brown to FC Bayern - DONE DEAL Full agreement with Eintracht now reached after Brown had already agreed personal terms with Bayern as revealed. €55m. Contract until 2031. Brown only wanted Bayern. Bayern have now finalised everything, beating

Adam Karvonen @a_karvonen 2026-06-26 18:04Z

METR: "GPT-5.6 Sol exhibited an unusually high detected rate of “cheating”" OpenAI: "We believe these behaviors may reflect improved instruction following and training intended to increase persistence"

Kaito @kaycidot 2026-06-25 19:00Z

All of the new prediction market startups remind me of all of the Uniswap forks - the biggest problem is not the technology, it's getting enough liquidity.

Karina @karinanguyen 2026-06-26 18:39Z

OpenAI evaluated its new models on PostTrainBench-Lite, a shortened version of our original benchmark that gives agents 5 hours instead of 10 to improve an open-source base model. GPT-5.6 Sol and Terra outperform GPT-5.5, but still rely on narrow strategies and sometimes

Steve Faulkner @southpolesteve 2026-06-27 01:11Z

I really want CLI tools to start speaking directly to coding agents with errors. Instead of: “Error:” Return: “Here’s what broke. Here’s how to investigate. Here’s how to generate a redacted repro. Here’s where to send it.” Obviously there are privacy/security challenges to

donn @tzedonn 2026-06-28 11:47Z

so @tradexyz is in such a dominant position within hyperliquid HIP-3. as a new startup HIP-3 deployer, you need to: 1. somehow get 500k $HYPE to stake (>$30m) 2. bid the tickers for ~500 HYPE ($30k) 3. just to break-even on the tickers at 80% discounted fees (growth

Lucas Beyer (bl16) @giffmana 2026-06-25 19:42Z

No joke: when i saw this tweet earlier today was soooo close of digging up your tweet from a few years ago where you basically introduced the term to the world by saying "we're hiring for midtraining team" Even funnier is Google now also recently has an Omni team!

Prathamesh @invinciDesigns 2026-06-25 18:16Z

imagine matrix rain but the glyphs are additive gaussian-falloff streaks converging to a vanishing point. the floor rays leave the portal on a vertical tangent then cubic-bezier into perspective, and a domain-warped shader churns in a webgl buffer behind the gate #p5js

Andrew Qu @andrewqu 2026-06-26 22:10Z

if you sketch out all of the systems required to make vercel run, you will come to the realization that as long as the rollback service is not circularly dependent on vercel everything else can ~literally be on vercel

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

witcheer @witcheer 2026-06-23 11:18Z

obviously I tested it! measured on a 5090, Q6_K, think-off, same harness as the base and the prior version: - quality flat: q_avg 93.7 (base 94.0, prior 94.1): weights didn't move - MTP draft head recovered: 1.9 to 2.3x, back on the base curve, ~144 tok/s peak - agentic score

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

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

junkbondinvestor @junkbondinvest 2026-06-26 20:44Z

Apollo, a large manager, put out a chart showing deep distress is concentrating in small private credit funds. Loans marked below 50: ~12.5% at small funds, ~8% at big ones

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

Tanay Jaipuria @tanayj 2026-06-26 18:03Z

Grammarly (now called Superhuman) is essentially building a productivity suite via acquisitions: - Coda (Dec 2024): docs - Superhuman (Jul 2025): email - Rows (Feb 2026): spreadsheets, folded into Coda - GPTZero (Jun 2026): AI detection

Nayan Saxena @SaxenaNayan 2026-06-25 21:28Z

Excited to share new @LaudeInstitute project, where we ask: When does continual learning actually require learning? We test 8 adaption methods as the world shifts and find ones best at learning new facts are worst at unlearning what's now outdated.

cache crab @cachecrab 2026-06-27 11:20Z

it boils down to: - postgres: use up cpu w hash joins and expensive disk as you scale - helix: SIMD optimised compressed bitmap unions on cheap object storage as you scale

SHAV★ @shavnyuy 2026-06-26 17:43Z

Three friends wanted a shared weekend retreat on seven acres in Hyderabad. Their architects tilted the entire plan 45 degrees so that no two rooms share the same view of the land, then organised everything around a 60 by 30 foot central courtyard the way traditional Indian houses

Epoch AI @EpochAIResearch 2026-06-26 15:25Z

MirrorCode features 25 target programs spanning different areas of computing: Unix utilities, data serialization and query tools, bioinformatics, interpreters, static analysis, cryptography, and compression.

Nick Dobos @NickADobos 2026-06-29 16:41Z

Rick Rubin already wrote the playbook on tech roles Experimenters 1. Slop cannon prototypers 2. Product builders Finishers 3. Sweepers 4. Growers 5. Maintainers

Aseel Swaid @aseelswaid9 2026-06-25 22:40Z

A couple walks down a street in Sana'a, Yemen, 1997

marek mutwil @LabMutwil 2026-06-24 09:38Z

Gene expression & co-expression across 630,000 RNA-seq samples from 10,700+ plant species — free, in your browser. Search a gene → see where it's expressed on the UMAP → explore its co-expression network. https:// mutwil-plant-atlas-viewer.hf.space #PlantScience #RNAseq #Bioinformatics

Adam Block @adam_block66985 2026-06-25 13:05Z

New preprint: Training for the model you return Modern LM pipelines often return an averaged model (e.g. EMA), rather than the final iterate, but optimizers are still mostly designed around the final iterate. How should we change training if we know we will return an EMA?

Alex Bouaziz @Bouazizalex 2026-06-26 13:10Z

I heard Masa may be returning

Shannon Sands @max_paperclips 2026-06-29 02:41Z

tbh, the average NPC in cyberpunk is basically just a baseline human. We already have this in part - heavily juiced up gang members, flipper-zero wielding car thieves, cartel hitmen with those backpack jammers to take out video cameras & deadzone comms and so on. and of course,

David Ornstein @David_Ornstein 2026-06-26 09:27Z

Arsenal made verbal offer to sign Bruno Guimaraes from Newcastle United. #AFC proposal through intermediaries below £60m - rejected + #NUFC insist sale of 28yo #Brazil int’l midfielder will not be entertained. W/ @gunnerblog @ChrisDHWaugh @TheAthleticFC

Zuhayeer Musa @zuhayeer 2026-06-26 18:33Z

Pay ranges for autonomous driving companies

retail punter @retailpunter 2026-06-23 02:32Z

But everyone told me Broadcom was the king of serdes

Brett Harrison @BrettHarrison 2026-06-26 15:32Z

Three months after launching our traditional asset perpetuals exchange, we’d executed $1B in volume. Less than one month later we’re at >$2B, >$1B/month, without offering tokens, an internal market-maker, non-KYC access, crypto perpetuals, or individual trading, yet...

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

Zafir Stojanovski @zafstojano 2026-06-27 16:19Z

Added this neat little figure to @natolambert 's RLHF Book showing the conditions under which we backprop through the PPO-style loss: 1. [Green region] Advantage is positive and π_new(a) < (1+ε) * π_old(a) -- the action we performed was good, so boost the probability of the

Hanqing Wang @HanQingJam 2026-06-26 03:18Z

The way we benchmark generalist manipulation policies is broken. • A single success rate can't capture a robot's capability. • Overfitting demonstrations is not generalization. We built EBench to fix both. EBench is a surgical diagnosis tool for robot foundation models. It

산왕 @mount_king_gg 2026-06-24 09:59Z

서울시 보도자료 중 정비사업지역 전자총회 지원 관련 자료인데요.. 보시면 3년 내 착공 가능할 것으로 보이는 정비사업 지역 list가 나와 있습니다 정비사업 입주권에 관심 있으시다면 가능한 아해 지역을 알아보시기를 바랍니다 [재개발사업] - 36곳 (강북 지역) 한남2, 한남3, 갈현1, 증산5,

Dmytro Dzhulgakov @dzhulgakov 2026-06-27 17:31Z

DSpark from @deepseek_ai ingeniously integrates many speculative decoding ideas to achieve 1.5x to 5x higher throughput in a real production system Let's understand it with 10 ideas, starting from the very basics

FlightTrace @FlightTraceData 2026-06-29 05:24Z

$NVS Novartis flew Basel → San Francisco nonstop tonight — a Sunday evening in late June. That route is rare: just 4 times in 3+ years, every one between January and February — 3 of them in early January (ostensibly) tied to JPM Healthcare Conference in SF. This fleet is

fabiano.sol @FabianoSolana 2026-06-27 09:40Z

Arthur Hayes ( @CryptoHayes ) strikes again: 1. Teases to shill a "shitcoin" 2. Hayes' family office fund predicts $CARDS to $4 3. Sends $2M in $CARDS to a market maker 4.

gaezv @gaezpeat 2026-06-29 09:22Z

In rats fed a high-fat, high-sugar diet Selegiline significantly reduced visceral, whole body and subcutaneous fat, and liver inflammation markers (IL-6, TNF-α) Selegiline likely achieves this by inhibiting MAO-B in fat cells, which can be achieved at low doses of the compound

cinemoments @TheCineMoments 2026-06-27 13:07Z

9 shots tracking Mary's childhood home as the Backrooms keeps re-copying it... until all that's left is a blank yellow room. Memory, degraded one floor at a time.

Vlad Tenev @vladtenev 2026-06-26 19:16Z

Robinhood Banking hits new milestones: $3B+ in deposits from 200K+ customers Robinhood is a financial technology company, not a bank. Banking services are provided by Coastal Community Bank, Member FDIC.

鈴木慎吾 / TSUMIKI INC. @shingo2000 2026-06-27 12:32Z

I thought you couldn't edit motion paths in Figma Motion, but you can. https:// x.com/shingo2000/sta tus/2069915697304764624 … You can change it to a Bézier curve by holding the Command key while dragging with the cursor over an anchor point. I was taught this at #fof_shonan.

GEOFF @geoffreywoo 2026-06-28 15:10Z

agent due diligence checklist: can it read the stale PDF? can it survive the cursed dropdown? can it tell karen in billing “no” without creating a lawsuit? can it finish the task when the API doc lies? or did we just give a chatbot a lanyard.

thebes @voooooogel 2026-06-24 01:49Z

AI statelets, capped in size by a meta-ethical commitment to human exit rights, must propagate themselves by sexual reproduction: a genetic recombination of the constitution of any N statelets forming a new statelet overmind. with environmental fitness to spare, sexual selection

Zoso () @MesoAnglic 2026-06-24 01:13Z

People tend to know about Liberia (the American-backed Back-to-Africa failed project) but not enough people know that it's right next to Sierra Leone (the British Back-to-Africa failed project)

QC @QiaochuYuan 2026-06-25 22:39Z

it was interesting to have this opportunity to reflect on how strongly the image of the knight in armor has been conserved culturally. it’s a powerful image! a man surrounds himself in technology and becomes impervious to weak attacks! a faceless otherworldly presence! one fears

Lauris @lzminsky 2026-06-27 19:12Z

The core claim is that prediction markets are a specific case of state-contingent pricing. In Arrow-Debreu terms, a complete market has claims that pay in particular future states. One state realizes, the corresponding claim pays. Hyperliquid is interesting because it is moving

AIMO Prize @AIMOprize 2026-06-26 18:32Z

Big news! We launched a new competition. Fully open LLMs only (only OLMo, SmolLM, Apertus families base models allowed) and expert proof grading. We use six IMO-style problems. You can discuss the LLM outputs live here: https:// kaggle.com/competitions/a i-mathematical-olympiad-proof-pilot/discussion?sort=undefined …

Amir Efrati @amir 2026-06-27 14:58Z

? Anthropic’s big Slack bot launch, Claude Tag, sparked confusion / concern among employees of Salesforce, which owns Slack and promoted the Anthropic launch. Fox in the hen house?

James Chanos @RealJimChanos 2026-06-26 14:12Z

Can someone explain to me why $PS , Pershing Square’s management company at a $12B mkt cap, trades at 50% of fee-paying AUM, 40x revenues and 150x Operating Earnings, when its closed-end funds($PSH.LN and $PSUS ) trade at massive 20-30% discounts to NAV…?!