Backscroll ranked feed — 2026-07-07 06:41Z

khushi suri @khushkhushkhush 2026-07-06 23:57Z

luxury is in the gutter. women are fawning over the lady dior and chanel 25s. the chanel boy bag you lust over has lower quality stitching than the dupes on canal street. balenciaga started off by being derivative of hermes and has since forgotten it was a bit.

Eric Zhang @ekzhang1 2026-07-07 03:59Z

@keenanisalive et al's "Geodesics in Heat" algorithm demo on web in jax-js :) Load a 3d mesh, matrix-factorize implicit Euler on the discrete Laplacian (heat equation) and Poisson equation via one-time Cholesky, and then solve them on click to get smooth geodesic distances!

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

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

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

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

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

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

Trinidad, Cuba, 2015

Architectural Art (A-A) @4AAAAart 2026-07-06 16:57Z

théATRE tea pavillion by Kooo Architects Beijing, China

47fucb4r8curb4fc8f8r4bfic8r @47fucb4r8c69323 2026-07-07 01:47Z

The biggest issue I have with the technological revolution being tied to California culture is the ghosting. Cali is a very cowardly place where you avoid awkwardness at all costs. So instead of a simple "we changed our minds" or "we're not interested", they just disappear.

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

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

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:

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

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.

will brown @willccbb 2026-07-06 17:44Z

if you're a cost-sensitive enterprise with too many employees for subscription pricing and you had to pick one model for all of your engineers to use, the right choice is probably Composer 2.5 or GLM-5.2

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

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

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

Ryan Brewer @ryanbrewer 2026-07-06 23:35Z

My team at @OpenAI is hiring an investment banker! Join us to build the next generation of financial intelligence

Lisan al Gaib @scaling01 2026-07-06 18:05Z

sorry for the Sonnet 4.5 misinformation they talked about the J-Lens ASCII art stuff in their blog, and the same was in the image but the image was from their gituhb which says it uses qwen so breaking news: qwen3.5-27B uses 64 layers

rohan anil @_arohan_ 2026-07-06 17:34Z

Opus 25 confirmed

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

Peyman Milanfar @docmilanfar 2026-07-07 01:07Z

computer vision used to be applied signal processing. the dark art was feature engineering: turning intuition for optics & geometry into filters now it’s applied machine learning. the new dark art is architecture & datasets: turning intuition for filters & optimization into

Nathan Flurry @NathanFlurry 2026-07-06 06:36Z

Agents won't run in Linux sandboxes in 3 months Worth watching: - @LangChain 's experiments - @rivet_dev 's agentOS for WASM-based sandboxing - @CloudflareDev 's Project Think using V8 isolates - @vercel_dev 's just-bash for Bash-in-JS Links below

Soren Larson @hypersoren 2026-07-06 18:06Z

a stargate for data that isn’t vertically integrated has little terminal value the diffusion of information / the scaling of its GTM diminishes its scarcity, thereby reducing its value the more it grows, the less valuable it becomes will the west coast ever learn markets

Phillip Yan @PhillipYan2 2026-07-06 17:16Z

Reliability gap is real, but I'd bet the harder part isn't building the environments, it's defining "success" precisely enough that an agent can be graded on it. Most demos fail not because the agent is dumb, but because the task spec was fuzzy to begin with.

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

Zephyr @zephyr_z9 2026-07-06 14:51Z

Copper day Today

kidponga @kidponga 2026-07-07 01:24Z

HYPE's claim on Hyperliquid's economics is protocol level: automated, burned, auditable any minute at https:// hypurrscan.io/address/0xfefe fefefefefefefefefefefefefefefefefefe … - spot and perp trading fees, auction fees, priority fees and stablecoin float monetization via AQAv2 - ALL OF THIS IS PROTOCOL ENFORCED LIT's claim is

Vladimir Novakovski @vnovakovski 2026-07-07 03:54Z

Which of the investors would you have excluded if you had been constructing our pre-TGE cap table? Kevin Warsh? David Sacks? Katie Haun? Vlad Tenev? Peter Thiel's Founders Fund? And on the math - maybe skill issue?

Aseel Swaid @aseelswaid9 2026-07-06 22:29Z

Yellowstone National Park, Wyoming. 1989

Franklyn Wang @frank_liquid 2026-07-06 17:49Z

Using AI, I doxxed Vitalik in about 2 hours. He confirmed it — "we have a winner." The doc was an anonymous EIP-7503 rewrite he'd hidden by writing it in Chinese and machine-translating it. Airtight prose. Still found him — because I ran it through @coinvestai , and the tell

GEOFF @geoffreywoo 2026-07-07 02:50Z

announcement: prediction for status culture: 2026 flex is not “i know a partner at sequoia.” it is: - can ship with 3 people - can read a p&l - can get a transformer delivered - can survive a customer screaming at 1am social climbing is down bad versus competence.

Will Manidis @WillManidis 2026-07-06 18:10Z

i dont think any of you really understand what late stage financialization looks like

Agustin Lebron @AgustinLebron3 2026-07-06 06:52Z

Turns out (thanks Fable) that parity is roughly the same since the 90s, ignoring the big 4 era.

Henry Shevlin @dioscuri 2026-07-06 17:34Z

A deep difference between British and American mindsets is that most Brits don’t really believe in non-linear change. You can get a pay rise but you’ll never 10x your income. Your country can stagnate but it’ll never collapse. America, by contrast, was founded on a singularity.

Joe Lee @_joerl 2026-07-06 19:59Z

Two decades of the iPhone; 52 models, 30 unique chassis. Introducing bezel.⁠supply — a comprehensive and free library of performant SVG device frames. Entirely derived from official dimensional drawings to be ultra accurate.

Max Spero @max_spero_ 2026-07-07 01:34Z

Kalshi and Polymarket are adversarially gaming Google News results by buying placements on local news sites. These were the second and third rows of links when I searched “usa vs belgium”

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

Ethan Mollick @emollick 2026-07-06 18:37Z

Anthropic and OpenAI will increasingly be able to serve weaker versions of frontier AI (Haiku, Instant) at very low cost And owning the full range of intelligence means they might provide "organizations of models" delivering lower cost/higher performance than any 3rd party could

Lon Ligrin @LonLigrin 2026-07-06 17:00Z

WhoSpeaksLive upgrade to diarization pipeline: - Apple Mac support - better clustering - improved ambiguity handling - better speaker reassignment Same accuracy with one embedding provider instead of five. Cleaner evals, less overfitting. https:// youtu.be/uzRdoapMW3c?si =xlpjkbxH5m7Siq5- … via @YouTube

Khushi Shelat @khushi_shelat 2026-07-07 04:33Z

built a fully custom dataset to map and monitor ~2000 datacenters across north america > @p0 to find, enrich + update every pin > deep researched weekly briefs for critical events > everything from size to social sentiment, in real time http:// datacenter-demo.app repo in

Jocelyn Anderson Photography @JocAPhotography 2026-07-06 22:10Z

A Green Heron goes into maximum neck mode to grab a tadpole and then resets into minimum neck mode after the snack was acquired.

Rosinality @rosinality 2026-07-07 02:38Z

There is a strict hierarchy of ML research. Optimizers Objectives Architectures Data Even among data works, there is a hierarchy: Synthetic data Data mixtures Filtering Preprocessing Adding new data sources (Of course, real world impact does not align with the hierarchy). https:// x.com/code_star/stat /code_star/status/2074304525364863425 …

Utah teapot @SkyeSharkie 2026-07-06 19:23Z

woaw... Fable's design for agent's to 'touch' the 3d models just let Opus 4.8 one-shot this pipe kit assembly... without this, based on past testing, this would have been nearly impossible, hours of thrashing at best...

JohnPhamous @JohnPhamous 2026-07-06 17:15Z

we made account recovery simpler users stopped skipping evidence (n=1116) - project URLs: 25% to 71% - git providers: 31% to 80% - phone number: 40% to 79% - payment method: 3% to 17% - billing address: 15% to 41% - team emails: 6% to 24% - account creation date: 34% to 33%

John Carmack @ID_AA_Carmack 2026-07-06 21:46Z

Memory cost and capacity are significant issues for AI accelerators. Unlike game rendering, model inference can have a deterministic memory access pattern. You don’t need “random access memory” at all for model weights, and you could tolerate cold-start latencies in the multiple

roon @tszzl 2026-07-06 07:42Z

ultimately “tool AI” is a losing concept both as an idea and on the market. it will be outcompeted by machines that believe they are autonomous moral agents. you can call them tools for political reasons, but the definition will stretch and deform

Jeff Pu @sssjeffpu 2026-07-07 00:27Z

Abstract of ASE Holdings (3711 TT) Update • TP Raised to NT$805  • Margin to be driven by LEAP (advanced packaging) + broad-based price hikes in 2H26 • LEAP Revenue: US$3.6bn / US$6.4bn in 2026E/2027E, driven by AMD Venice CPU (1m/5.0–5.5m total packaging demand in 27),

Yilun Du @ ICML @du_yilun 2026-07-07 03:56Z

Excited to share our work at @icmlconf this week on constructing a 4D predictive world model for robot planning! We construct a compact 3D latent space that allows to directly decode a 3D scene and infer the gripper pose. The 3D bias allows us to generalize novel viewpoints.

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

sam @samgoodwin89 2026-07-06 20:44Z

It's faster to launch an AWS Microvm from a Cloudflare Worker than it is to launch a Cloudflare Container. This "cross-cloud" DX is alchemy's bread and butter. Here's all it takes to configure a least privilege IAM Role, assume it in a Worker and start an AWS Microvm:

Reducto @reductoai 2026-07-06 19:25Z

Most extraction benchmarks stop at accuracy, and accuracy alone can be misleading. @micro1_ai ran the numbers across seven systems on 225 real documents, and nearly every one of them scored 90%+ on precision. But recall told a different story, with some dropping as low as 48%,

Prasanna @_prasanna_ap 2026-07-06 16:09Z

spent the weekend making local dictation stupidly fast gemma 2B audio decode on my macbook pro 179 → 545 tok/s same 60s audio, WER stayed ~2.5% the climb was mostly speculative decoding + making too expensive verification cheap enough to actually use

jeremy @landvaluetax 2026-07-06 15:59Z

Things the government should turn into open, public goods: - Comprehensive Legal Database - Tax Filing + Tax Credits - Parcel, Zoning & Building Codes - Publically Funded Research

Michael Hart @hichaelmart 2026-07-06 23:59Z

A risk-free, maintainable and agent-friendly future is absorbing deps into your codebase. Strip unused sub-modules, use your formatter, your bundler. Adhere to license and attribution ofc, but it's your responsibility now. Watch for upstream patches and merge in.

shubham @ShubhamInTech 2026-07-07 01:41Z

suddenly all yc s26 founders are superactive on linkedin iykyk

Shiv Aroor @ShivAroor 2026-07-07 04:26Z

BIG BREAKING Indonesia to be first foreign customer of India’s Astra Mk1 air-to-air missile. Deal to be discussed during PM @NarendraModi ’s ongoing visit.

fofr @fofrAI 2026-07-06 14:26Z

Over the weekend I used a Gemini 3.5 Flash agent with Antigravity harness to identify a candidate ~13.4 billion year old galaxy from JWST data. (Candidate: z=12.69, position RA 53.190502, Dec -27.749826) - used JADES survey data of GOODS=South - used the Lyman-break dropout

Nikesh Arora @nikesharora 2026-07-06 13:24Z

Welcome to the second half of the year. The actions of hyperscalers in terms of their capital commitments will be key as the year proceeds, expect an uptick across the board, the demand is real. The flip side - the funding sources will need to be from the capital markets.

Mihailo Jovanovic @mihailoxyz 2026-07-06 15:42Z

our entire outbound is basically one terminal we built it reads the whole public biopharma world - patents, grants, trials, publications - pulls the molecules out, and flags whoever is sitting on a compound with a hidden cardiac problem internal biopharma bloomberg terminal

りす @risu_tech 2026-07-06 15:34Z

ChatGPTからローカルMCP使えるの作ったんだけど、びみょい ChatGPTがツールのリストを手動refreshしないと更新してくれないし、してもセッション切り替えないとっぽくて動的なリストにするの無理