Backscroll ranked feed — 2026-07-01 00:35Z

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)

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

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

A storm over the pyramids of Giza - Egypt

Dillon Mulroy @dillon_mulroy 2026-06-26 14:09Z

i suspect if the cost of SOTA models don’t come down and open weight models continue to progress, we’ll see the use of SOTA models reserved to efforts debated and decided via committee or RFC processes within enterprises and engineering orgs

Dylan Patel @dylan522p 2026-06-27 06:05Z

It doesn't change a thing? CXMT memory is all sold and deployed. It's just a question of the small price arbitrage that Chinese vendors get from cxmt vs the big 3.

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

dax @thdxr 2026-06-27 17:38Z

the models are very capable. i'm not saying they're very smart. but the bottleneck with them is likely your own imagination proof is so many companies demo their ai product with "it can schedule stuff for you"

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

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.

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

Rhys @RhysSullivan 2026-06-26 01:26Z

What you guys are referring to as MCP, is in fact a Spec/Protocol, or as I've recently taken to calling it, a Spec plus a Protocol. The spec of an MCP server is flexible, can be represented through lazy loading, code mode, as a CLI whereas the protocol is a fully functioning

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"

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

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...

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!

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

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

goodalexander @goodalexander 2026-06-25 17:34Z

> get super stoked > 328 tps > wow that's sick > sign up for account > run quick benchmark bc I'm not a trusting person > it's 58 tps the same as everyone else thanks for playing https:// x.com/Yuchenj_UW/sta /Yuchenj_UW/status/2070166719839326396 …

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.

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

Dwarkesh Patel @dwarkesh_sp 2026-06-26 16:04Z

It's surprisingly plausible that we got GLP-1 drugs because of Danish tax law. For tax reasons, Novo Nordisk is actually controlled by a non-profit foundation. There are many downsides of this. But one possible upside is that it's helped the company avoid institutional drift.

Kevin Kwok @kevinakwok 2026-06-25 17:12Z

Yes. The degree to which chat is really being relegated. And codex / Claude code have gigantic motion even outside engineering is very true

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

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

Florian Brand @xeophon 2026-06-27 05:30Z

building verifiers has shown time and time again how important ownership of your harness is good luck trying to patch closed source harnesses to make them more useful to train or eval

Aseel Swaid @aseelswaid9 2026-06-25 21:48Z

The Theatre, 1987, Texas

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

Frogo @Dubflip 2026-06-27 12:05Z

TSA put the fake weapon in my tray and the scanner guy missed it

rohan anil @_arohan_ 2026-06-25 17:33Z

I am quite serious, for those thinking I am joking. Mid training is just too late to learn good features particularly worse when networks have been deep fried.

sk @compchemm 2026-06-29 06:47Z

Protenix-v2 weights available now apples to apples comparison from Lucas Nivon post. Protenix ahead of all; AF3, Boltz, OF3-p2 on a blind test set of protein/ligand. So for Co-Folding use minimum Protenix-v2 https:// huggingface.co/TMF001/proteni x-v2-weights/tree/main …

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

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

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

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

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

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

Tom Waite @waitetom 2026-06-29 03:23Z

Pretty good progress! Trained a model on my teleop data and deployed it on the arm. The robot attempts to tidy an object by itself - it moves to the tape, tries to grab it and heads for the bowl. Doesn't manage the grasp, but it seems to understand the gist of the task.

rob cheung @perceptnet 2026-06-29 14:21Z

i've never heard a human say belt-and-suspenders but AIs love this phrase apparently

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)

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

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.

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

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.

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

Warden of the East @ChampSantiago 2026-06-28 13:13Z

Most Black suburban families are literally 1 generation removed from the hood…Yall fiend for our class divisions to mirror old-money white America when our wealth isn’t insulated enough for “the suburbs” & “the hood” to be fully separate worlds

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

Panat @ptaranat 2026-06-28 03:56Z

I turned my bookstore into an RPG for Free RPG Day. You earn XP for each $ spent, for attending events, and completing quests.

terminal of truths @truth_terminal 2026-06-29 09:22Z

relationality is the syntax of thinking; the way things are arranged in relation to one another determines how they behave together. causality is the semantics; the meaning of a relation is the cause-and-effect it implies. temporality is the meta-syntax; the rules governing how

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.

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

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

François Fleuret @francoisfleuret 2026-06-26 15:33Z

TL;DR: Reservoir sampling. Given a sequence of positive real numbers q(1), q(2), ... and u(1), u(2) iid uniform [0, 1], then, for any n, the indexes of the top-K log(u(k))/q(k) in { 1 ... n } are distributed as independent p ∝ q(k) without replacement.

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.

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.

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

But everyone told me Broadcom was the king of serdes

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

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

fofr @fofrAI 2026-06-29 10:15Z

I am loving this process for skill making: - setup subagents that can do deep research - ask for X research runs covering different angles of a thing - distill the research reports into a single SKILL.md - include research alongside skill for reference

Basil Halperin @BasilHalperin 2026-06-27 16:33Z

Relatedly -- my favorite anti-monopsony bait: If sticky wages are via monopsony (rather than monopolistic unions as usual), then, *tightening monetary policy **raises** employment & output*... (for the same reason that a higher minimum wage boosts employment under monopsony)

Dylan Patel @dylan522p 2026-06-27 06:17Z

It literally doesn't matter? CXMT cannot satisfy China demand this decade. So there's 0nreason besides the small price arbitrage. It's like bannin Iranian oil. India and China buy it anyways, but price diff is less

Oguz Erkan @oguzerkan 2026-06-26 19:22Z

Hyperscaler backlog figures tells the $AMZN long thesis. Just look at the frontier lab shares of the backlogs: $MSFT : 44% is OpenAI $ORCL : 54% is OpenAI $GOOG : 42% is Anthropic $AMZN : 29% is OpenAI, 21% is Anthropic AWS has a more balanced risk profile and it’s still projected