Backscroll ranked feed — 2026-07-06 23:30Z

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

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

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

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:

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

Phillip Yan @PhillipYan2 2026-07-06 17:13Z

Kitchen analogy tracks but the hard part is the expediter role: someone (or something) has to catch when the fast-worker shipped something the planner didn't actually ask for. Multi-model setups fail more on handoff drift than on any single model being weak.

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

Opus 25 confirmed

47fucb4r8curb4fc8f8r4bfic8r @47fucb4r8c69323 2026-07-06 11:50Z

There are higher order effects to this. If weird appears to get the money, strivers will affect weirdness even if they're not weird. And you'll see effects in the language, like "normal" will be used to mean undesirable in many cases and, if the trend last long enough, normal

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

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

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

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

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

Aseel Swaid @aseelswaid9 2026-07-06 22:00Z

Across the divide, 1984

Coframe @coframe_ai 2026-07-06 22:34Z

We drove 410% conversion lift on @Replit 's enterprise funnel, 2x-ing the number of demo requests. All in a matter of months. "Coframe makes you feel like you have a team of 100 people. There's nothing on the market like it." ⧉ x ⠕

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

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.

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

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

Zephyr @zephyr_z9 2026-07-06 09:52Z

Observe the ASP growth since Feb

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.

Kit Langton @kitlangton 2026-07-06 20:30Z

I packaged up a little @EffectTS_ -based HTTP cassette recorder library we made for testing slow, expensive, & flaky LLM calls in @opencode . Here's a demo.

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.

Easy @NotSoEasyMoney 2026-07-06 20:45Z

So wait… Bonk dao wasn’t a hack? It was a giga whale buying enough supply to control the vote entirely? And voted to send themself $20mil… And the vote was live for 6 days… and no one noticed so it just went through…?

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

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

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

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

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

Jan Kulveit @ ICML @jankulveit 2026-07-06 15:56Z

Eric Drexler was mostly right about ecosystems (as opposed to MIRI central views) and mostly wrong about "tools". The problem is 'agents' are a highly convergent solution. Evolution also does not somehow intrinsically want agents: genes want a tool, a design stance system, to

Matt Rickard @mattrickard 2026-07-06 16:08Z

At corigin internally we have extremely strict, opinionated, and extensive post-agent code quality gates. There’s so many custom AST rules and autofixes it’s not really a “linter” anymore. Every agent code smell gets a regression AST rule. Some can be fixed deterministically,

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

The Icahnist @TheIcahnist 2026-07-06 22:06Z

Invisible companies are the pathway to empires: Someone rolls up a boring, stigmatized industry, builds cash flow, then uses that capital to buy trophy assets that elevate their status. Pinault rolled up small timber firms, then used the cash flow to buy Gucci and Balenciaga

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%

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:

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

Paul Benigeri @benigeri 2026-07-06 18:25Z

Usage rights are the most mispriced line item in creator deals. A creator charging five figures per reel told me he sells 30-day usage windows and never checks what runs after day 30. He also has no idea what brands spend behind the post. Some companies pay $5,000 for the

Artificial Analysis @ArtificialAnlys 2026-07-06 18:12Z

AutomationBench-AA evaluates models on 657 workflow automation tasks developed by Zapier from real workflow patterns on its platform. Models orchestrate work across 40 simulated SaaS app environments via REST APIs, and every task is graded programmatically on the final state of

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

SGLang @sgl_project 2026-07-06 17:23Z

DSpark marks a new era where spec decoding no longer slows down under high concurrency. We took time to carefully optimize variable-length verification, keeping it fast while maintaining high verification quality. Run Qwen3 and DeepSeek-V4 with DSpark in SGLang now, let's

Nicholas Charriere @nichochar 2026-07-06 18:00Z

we're very lucky that codex is open source! here's a particularly interesting part of the codebase: the prompt around the /goal feature (how codex decides to keep going or stop):

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

Alex Smith @ninja_maths 2026-07-06 17:06Z

For anyone wondering how a third-grader can complete six years' worth of math in a single year AND score a 5 on the AP Calculus exam. This knowledge graph spans 3,000 math topics, from 4th grade to the university level, providing the perfect basis for mastery learning.

anu @anuatluru 2026-07-06 22:00Z

Be honest about your particular set of interests + skills and clever people will figure out how to leverage them instead of just trying to fit you to a type. Even more important now that traditional jobs/functions/roles are becoming archaic. Claim your own zone of genius.

Orson @orsonwbs 2026-07-06 16:49Z

New white hat bizopp offer that we've been printing with AIMA, $90+ payout, $37 CTC. White hat and FTC compliant The ads are stupid easy to make. Ugly statics with basic copy Multiple affiliates already running this and seeing conversions. If u want access apply at

Mitchell Hashimoto @mitchellh 2026-07-06 16:59Z

One of the changes made to improve Ghostty IO throughput was to introduce a new IO gather thread, bringing our total thread count to 3 per terminal instance. Under heavy IO load, this keeps our VT processing working at nearly full bandwidth without stalls. Previously, Ghostty's

seb @hiiinternet 2026-07-06 18:38Z

In the last 4 weeks 168 engineers got 477 interview requests from startups via Fonzi with an average bid of $203k and a high of $350k (base salaries) Fonzi is an ai recruiting agent that lives in iMessage, intros you to startups looking to hire you with salary bids upfront,

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

Omar Rizwan @rsnous 2026-07-06 16:56Z

Physical scale that can weigh a website

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

Trinidad, Cuba, 2015

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

USD.AI @USDai_Official 2026-07-06 15:30Z

New loan executed: $11.3M 32 B200 Blackwell servers (256 GPUs) are now online in Texas. Capital has been released from escrow to the borrower, pushing the yield from 7% to 11.5% APY.

Judge Holden @the_judge1111 2026-07-06 20:16Z

not sure this is a good take accel distributed in kind (verifiable via form 4) everyone participated in upside (if they held) accel derisked their fund bigly not their job to make buy sell decisions for a public position per se; curioius @endowment_eddie take

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

Natasha Mascarenhas @nmasc_ 2026-07-06 16:06Z

Top AI startups continue to line up tender offers at a regular clip. The latest? ElevenLabs, which is discussing an employee tender that could take place in September. Scoop w/ @yazhous

Carl Bovis @CarlBovisNature 2026-07-06 10:34Z

A gorgeous Long Tailed Tit in a rare moment of stillness.

Polymarket @Polymarket 2026-07-06 16:04Z

JUST IN: Florida now accounts for 1 in 7 homes for sale in America, with nearly 45% of Florida listings taking price cuts — Parcl

Alex Kesin @alexkesin 2026-07-06 21:47Z

Yes I find that number insane as well. CAH rated by UBS w/o risk adjustment @ $2.5 - 3B, and acromegaly market on somatostatin is <$1B...

Baraq Lipshitz @baraqlipshitz 2026-07-06 17:23Z

Project Update: I’m training a World Action Model for general bimanual manipulation. Here’s everything I’ve learned so far, including the CPU contention bug that was silently corrupting my data. Thanks to @nvidia , @AWS for the compute and @ethroboticsclub for the hardware! (1/8)