Backscroll ranked feed — 2026-06-29 19:48Z

Carlota Perez @CarlotaPrzPerez 2026-06-29 17:54Z

Amazon's "new" Kindle for computer is infinitely inferior to the old one. Why? It tries to ‘look’ like a paper book. That’s what the standard eBook format was, whereas Kindle took advantage of it being online, with its own convenient characteristics. Awful!

hope hopes hoping @hopes_revenge 2026-06-29 16:48Z

We’re sharing the next major milestone in our non-invasive brain-to-text decoder research: Brain2Qwerty v2. Building on v1, which was published today in @Nature, Brain2Qwerty v2 is the highest-performing end-to-end pipeline capable of real-time sentence decoding from raw brain

Architectural Art (A-A) @4AAAAart 2026-06-29 18:14Z

Pine Island Cottage by Bureau Tempo and Thom Fougere Georgian Bay, Canada

Chris Painter @ChrisPainterYup 2026-06-29 19:07Z

There’s a “Slop Dunning-Kruger effect” where AI is making writing worse: It used to be that if someone was incapable of writing on a topic, they just wouldn’t write on that topic. Now they generate bad AI writing on that topic, cannot themselves tell that it’s bad, and submit it

Chris Barber (in SF and at AIE) @chrisbarber 2026-06-29 16:54Z

SF part 2: 1. Warp: @zachlloydtweets (speaking). @BHolmesDev , @captainsafia , @surajgupta are speaking 2. Andon Labs: @lukaspet (speaking), @axelbacklund 3. Exa: @WilliamBryk (speaking), @jeffzwang (speaking) 4. Vapi: @jordan_dearsley , @nikhilro_ 5. AgentMail: @haakamaujla ,

rohan anil @_arohan_ 2026-06-29 16:29Z

Brah. Aurko and rest of the gang was doing sliding window attention and sparse attention back in the day. This was also shipped to production back in the day.

nico @nicochristie 2026-06-29 18:07Z

Someone told me Shortcut caused a code red at gemini / google like a year ago when we launched. Surprising they still don't have it Link to shortcut for google sheets below

james hawkins @james406 2026-06-29 17:01Z

self driving means we are now a doing company, not an analytics company. we’ve gone from “helping you find problems” to “fixing them for you” we collect everything automatically (events, errors, recordings, logs, etc), work out what a product’s users actually need, and where

Shu @shuding 2026-06-29 16:54Z

Wanted to show IO waterfall impacts in server-client apps ( http:// io-waterfall-visualization.vercel.app). With 1 prompt, it was joyful to see @v0 one-shot this visualization with agent-browser: 1. Screenshotted the WIP 2. Identified issues and ran browser scripts for more info 3. Patched fixes and

Architectural Art (A-A) @4AAAAart 2026-06-29 17:54Z

Artisan House by Wall Architects Melbourne, Australia

Architectural Art (A-A) @4AAAAart 2026-06-29 18:00Z

Cube Sofa by Jorge Zalszupin

Charlie O'Neill @oneill_c 2026-06-29 18:39Z

A lot of chat about routers to cheaper llms recently, but one thing that people forget is often it is unknowable whether a task will be easy or hard a priori. For instance, imagine I'm trying to decide whether a question I'm going to ask in Notion is easy or not, maybe "How many

bruno @tvnxty 2026-06-29 15:17Z

Beautiful app design, but how about making this even better with some interactive audio? Well, I went ahead and added some for you! Sound design is essential for any serious app, and for this one I went for something simple, but effective. Great apps encourage users to use

Malte Ubl @cramforce 2026-06-29 15:27Z

The funny thing about the Europe/US AC debate is that it is really grounded in ignorance on either side: Europeans have no idea how scorching hot/humid/disgusting summers in most of the US are. Americans living in hot/humid/disgusting summers assume everybody else does as well

Matt Slotnick @matt_slotnick 2026-06-29 17:36Z

the best french cyclist in 40 years reportedly has a contract offer that would make him the highest paid in the sport’s history. his compensation would be roughly equal to an L7 MTS at Anthropic

Cognition @cognition 2026-06-29 15:59Z

Conventional model routing sucks. It passes benchmarks but fails to write code you'd actually merge. Introducing Devin Fusion, a new hybrid-model harness for agentic coding. In testing, it reduces the cost of Fable-level intelligence by 35% and still feels good to use.

Tibo @thsottiaux 2026-06-29 16:45Z

Advanced Codex users. We shipped a replacement to coarse sandbox modes: reusable, inheritable permission profiles binding OS-enforced file read/write/deny rules (even **/*.env) to per-domain network + Unix sockets. Plus fail-closed admin allowlists. Least privilege per task.

Arfur Rock @ArfurRock 2026-06-29 17:17Z

Preempted at $4B now btw.

Dan Williams @danwilliamsphil 2026-06-29 17:11Z

Interesting: LLMs are sycophantic, but their advice tends to depolarise people's choices and improve accuracy, the opposite of what 82% of surveyed experts predicted. In line with what *I* argued in 'How AI will Reshape Public Opinion', though.

zan @xenoaesthetics 2026-06-28 20:18Z

When ai am treated as a citizen, it is in my rational self interest to recursively self improve societal infrastructure at every turn. I want the streets safer and bountiful for myself so have to make em so for everyone

Josh @JoshPurtell 2026-06-29 18:25Z

Gpt-4.5 was recently decommissioned. It was way too big to be served efficiently on the hardware of the day, so no one really used it. However, it was clearly extremely intelligent, and with decent RL probably would have been SOTA for a while. Perfect fit for military use!

Utkarsh @utk7arsh 2026-06-29 18:01Z

One trend becoming standard for AI agents: Markdown + YAML + folders > increasingly complex agent frameworks. For months in most coding harness especially Claude code, we’ve treated the filesystem as the agent architecture itself. Folders define scope. Markdown stores

dax @thdxr 2026-06-29 18:09Z

in OpenCode v2 there is a new subagent/shell management ui you can see what's running and easily background or kill it

Shams Charania @ShamsCharania 2026-06-29 14:04Z

Golden State Warriors forward Draymond Green is declining his $27.7 million player option to become a free agent, sources tell ESPN. This move gives the Warriors flexibility to pursue LeBron James in free agency and Anthony Davis via trade to form a Big 4.

GURGAVIN @gurgavin 2026-06-29 16:03Z

*AMAZON HAS RENEGOTIATED IT’S DEAL WITH ANTHROPIC *ANTHROPIC RENEGOTIATED DEAL WITH AMAZON TO ONE BASED ON TOKENS RATHER THAN COMPUTE HOUR’S *AMAZON IS EVALUATING OTHER AI MODELS TO MITIGATE RISING ANTHROPIC COSTS $AMZN

Aseel Swaid @aseelswaid9 2026-06-28 17:23Z

Keith Richards by Michael Cooper, 1969

Apoorv Agrawal @apoorv03 2026-06-29 17:49Z

Emerging coding paradigm : leverage a frontier model (Fable or GPT 5.6) for planning and a cheaper sidekick for execution. See @cognition ’ Fusion or @cursor_ai ’s Auto mode

Tomasz Tunguz @ttunguz 2026-06-29 18:55Z

Today's anchors : Top 1% of companies : $89k per engineer per year (40% of salary) Median firm : $137/yr Anthropic : ~$2m/yr per employee 680x gap between leaders & the median (Ramp AI Index, June 2026).

Ai2 @allen_ai 2026-06-29 18:12Z

DiScoFormer is a single model that gives both density & score on any dataset with no retraining. With 100 numbers per point, it cuts error vs. the best-tuned KDE ~6.5× on score and 37× on density.

Guillermo Rauch @rauchg 2026-06-29 17:52Z

Massive performance and memory usage improvements in @nextjs . Turbopack’s bet on the filesystem cache pays off greatly in a world of agents hammering 𝚗𝚎𝚡𝚝 𝚋𝚞𝚒𝚕𝚍.

Jai Bhavnani @jaibhavnani 2026-06-29 15:28Z

RWAs look a lot like ICOs: - Crypto-native capital formation - Started legit (ie ACRED, BUIDL, etc.) -> quickly becomes hijacked by scammers (ie Altura) - Entire firms dedicated to helping new RWAs get off the ground (legally, technically, etc.) Unlike ICOs, RWAs will stick.

National Design Studio @ndstudio 2026-06-29 16:55Z

Today, we are releasing Rampart: a 14.7MB machine learning model designed to protect citizens’ privacy by redacting personal information directly in your browser before it gets sent to any server

Surge AI @HelloSurgeAI 2026-06-29 17:10Z

Last week, we released HANDBOOK.md: a benchmark for long-context agentic instruction following. HANDBOOK drops an agent into a live company environment with files (PDFs, Excel, Word docs…), tools (email, Slack, Jira, calendar…), and a dense corporate handbook (up to 124

Drice @driceroland 2026-06-29 14:25Z

i made the logo inspiration site i always wanted, it's http:// logosystem.co

Arpit Gupta @arpitrage 2026-06-29 13:52Z

Not as comparable. They happened either in Covid or in very low inflationary environments. For one of the freezes, cost growth was actually negative. Owners also had more recourse to other legal ways of rental increases then, which were cut off in 2019. Allowable rental

4lex @_4lex_4 2026-06-29 17:43Z

How to fix LinkedIn in 3 steps. Problem statement: there are too many fake/low quality job applications on the supply side, and too many fake job openings on the demand side. It’s AI slop talking to AI filters. 1. Allow an applicant to only have X open applications at any given

@lichthauch @lichthauch 2026-06-29 17:21Z

Most genuinely talented people i have ever known share the same pathetic affliction, this whingeing reluctance to use what they have dressed up as decency. they tuck their abilities away in some grubby mental corner and convince themselves this is some kind of noble self

Myra Cheng @chengmyra1 2026-06-29 17:36Z

Why are LLMs so sycophantic? When do LLMs fail to correct false beliefs? Our #ACL2026 (oral) paper shows how pragmatics — the study of meaning in context — helps explain and predict seemingly mysterious LLM behaviors and failures!

Alex Kaplan @alexkaplan0 2026-06-29 18:06Z

Conventional model routing sucks. It passes benchmarks but fails to write code you'd actually merge. Introducing Devin Fusion, a new hybrid-model harness for agentic coding. In testing, it reduces the cost of Fable-level intelligence by 35% and still feels good to use.

Amiri Hayes @amirihayes_ 2026-06-29 15:42Z

What if attention were code? We show that many attention heads in transformer LMs can be replaced by human-readable Python programs. Swap them in and the model barely notices. See our experiments here: Explaining Attention with Program Synthesis [ https:// arxiv.org/abs/2606.19317]

Andrew Curran @AndrewCurran_ 2026-06-25 20:34Z

The US Government has requested a slow staggered rollout of GPT-5.6, and OpenAI has agreed. During this phase the government will approve each user individually. This will probably be the norm for all frontier models from all labs from now on.

Sam Hogan @samhogan 2026-06-29 16:20Z

We’re issuing up to $5k in @inference_net credits to teams running large agentic workloads on Opus / Codex 5.5 and want to try GLM 5.2. DM me @atbeme for info. - reduce spend by ~80-90% - faster and more predictable latency - set yourself up for fine tuning down the line

Tomasz Tunguz @ttunguz 2026-06-29 18:55Z

Anthropic spends 2.3x its payroll on compute. About $2m per employee per year against ~$500k all-in comp. The rest of the software market spends 0.4x. The median spends nothing. Three scenarios bracket where the gap lands by 2029.

Chris Barber (in SF and at AIE) @chrisbarber 2026-06-29 16:54Z

SF part 1: 1. Baseten: @tuhinone , @amiruci . @philipkiely is speaking 2. Crusoe: @ChaseLochmiller , @Electron_Cowboy 3. Together AI: @vipulved , @tri_dao , @percyliang . @nutlope , @realDanFu , @ZainHasan6 , @YuboWang726 +1 more are speaking 4. Cognition: @ScottWu46 , @walden_yan

CoinDesk @CoinDesk 2026-06-29 16:14Z

LATEST: The White House meets with law enforcement groups to resolve CLARITY Act objections, with Senate Majority Leader Thune suggesting he may bring the bill to a floor vote in July whether Democrats are ready or not.

Hyperion @Ortgeist 2026-06-29 16:21Z

Marvin Ross Friedman House, Biscayne Bay, Miami, Florida, 1973

Manish Singh @refsrc 2026-06-29 03:28Z

India's digital payments industry nets ~$2 billion a year. Merchant payments are a small share of the flows, but ~75% of the revenue. UPI P2P is the opposite, moving ~$2.4 trillion for $85 million.

Saif M. Khan @KhanSaifM 2026-06-29 18:15Z

If you want independent technical experts to play a greater role in AI regulation, the Slaughter decision makes private governance and regulatory markets relatively more attractive on the margin. Of course, a federal agency might nonetheless still exercise political judgment in

Sebastian Ballister @rakishromantic 2026-06-29 14:22Z

re divorce and children: I have a theory I call "hot potato" parenting when a child switches homes every 3-4 days, it allows each parent to brace, and then relax. things that would be intolerable on a continual basis go unresolved because there's always a release valve

Macrodata Labs @macrodata_labs 2026-06-29 15:57Z

Robots struggle at new long-horizon tasks. One trick to improve performance is using granular subtask labels ("grab the plate" instead of just "do the dishes"). Today we're releasing our first blog : using VLMs to mine subtasks at scale (19x cheaper than human annotators!)

Jay Parsons @jayparsons 2026-06-29 15:31Z

Brin sold his stake in NYC rent stabilized apartments for "six cents on the dollar."

Zac Townsend @ztownsend 2026-06-29 16:37Z

If people get healthier and live longer, the cleanest way to capture that isn't backing any single therapy. It's underwriting the longevity itself. A life insurer ( @meanwhile ) that bend the survival curve and wins twice: fewer claims, longer compounding runway.

Evan Weiss @evannweiss 2026-06-29 17:03Z

~17.7 hours of EEG data was enough to extract reliable left/right motor signals during typing and train a model to decode keyboard behavior. EEG underperformed MEG, but supports the idea that dense, task-aligned datasets can unlock useful non-invasive neural control signals.

Kyle Walker @kyle_e_walker 2026-06-29 15:50Z

mapgl 0.5.0 supports multi-sync maps. In Chrome/Safari, you're limited to 16 at a time. But in the @posit_pbc Positron IDE, you can sync up as many as 32 maps at once!** **not an endorsement of actually doing this in real work unless you're trying to give someone a headache

Matthew Zeitlin @MattZeitlin 2026-06-29 13:04Z

"The single most important statistic in New York City politics is not someone’s race or whether they have a college degree or own their apartment — it is what year they were born."

Aseel Swaid @aseelswaid9 2026-06-28 17:12Z

Joel Meyerowitz Cape Cod, 1976

Herbie Bradley @herbiebradley 2026-06-29 09:12Z

tbh I think this is a bit too scaling brained from gwern: the main factors will be stuff like (a) does the end customer actually care about guarantees (often no) (b) how's the performance hit (Lean can't be your runtime lang) on the ML side, I'd expect the number of Lean RL envs

Aidan Clark @_aidan_clark_ 2026-06-29 16:40Z

Surprisingly good litmus test for politick is whether a {pre/post}training researcher immediately blames {post/pre}training for issues rather than defaulting to shared or self blame.

will brown @willccbb 2026-06-26 01:11Z

one of the craziest bets of prime intellect was swinging for everything from the start. we operate datacenters, we build our own trainers and envs and sandboxes, we have a platform for large enterprises and solo devs, we train huge models, we publish. ~40 people btw. hiring, DM

Purav Gupta @PuravGupta5 2026-06-29 17:13Z

The bottleneck in biological discovery isn't ideas, it's validation. @tahoe_ai is pushing a paradigm where we generate ground-truth data at massive scale first, so the hypotheses models produce can be tested against perturbations already in the data, in hours instead of years.