Backscroll ranked feed — 2026-06-30 02:32Z

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

Artisan House by Wall Architects Melbourne, Australia

Jukan @ICML @jukan05 2026-06-29 22:10Z

Rubin Ultra won’t be able to gain an edge over MI500.

Jack @tracewoodgrains 2026-06-29 18:19Z

it is strategically and morally bankrupt to work to drive everyone who rejects the Omnicause away from supporting queer people. people with this attitude are unambiguously working to undermine gay and trans rights and they should be responded to accordingly

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

John Arnold @johnarnold 2026-06-30 00:24Z

Forget trust stacking. The tax break for founding and investing in tech startups shouldn't exist at all. Capital already floods into venture without QSBS. Exempting up to $15 million in capital gain is a windfall for investors to do something they were already eager to do.

Intelligent Internet @ii_posts 2026-06-29 19:15Z

You don't need Fable for the most complex tasks, from training models for protein prediction to optimising compilers Our open source Zenith harness takes base models to the top of FrontierSWE via adaptive self improvement GLM 5.2 next https:// ii.inc/blog/post/zeni th …

Jukan @ICML @jukan05 2026-06-29 23:58Z

KIS: On Memory LTAs With suppliers' commodity DRAM operating margins already at 80% and gross margins above 90%, prices cannot be raised endlessly on supply and demand dynamics alone. What matters is that at price levels where supplier operating margins exceed 80%, it was the

Corey Hoffstein @choffstein 2026-06-29 23:59Z

I know it's kind of stupid, but when something is 20% above its 200-day moving average, you'd need to see ~61% annualized return to maintain that distance. g ≈ 2Y / (N-1) (g is the daily growth rate; Y is the distance above moving average; N is the length of the moving

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

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

Seth Bannon @sethbannon 2026-06-29 22:42Z

A blood test just beat the scan. BillionToOne’s test predicted survival in cancer patients on immunotherapy better than standard-of-care imaging. 12 tumor types, 142 patients, progression flagged ~62 days earlier. Imaging has guided oncology for decades. This is a big deal!

Kevin A. Bryan @Afinetheorem 2026-06-29 18:52Z

The population of SF has gone up 6% over the last *25 years* despite a massive boom. 6! Alameda: 13.6%, Marin 2.6%, San Mateo 5.2%, Santa Clara 13.7%. US as a whole: 22.3%. MAINE grew faster than the whole SF metro. NEWFOUNDLAND grew faster than SF since 2000! Utter insanity 1/2

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

Michael @endpointarena 2026-06-29 21:17Z

Endpoint Arena is building a mobile app called Anecdote Bio. The problem with AI is that it rehashes the same public information to predict trial outcomes. The real alpha is data in your phone. Here is how it will work and how you can make money.

Max Prokopp @maxprokopp 2026-06-29 18:12Z

Handheld camera motion for Seedance 2.0 I hate blender so I stage it on my iphone. You stage the scene (open source link below) and can walk around it and capture real camera movement in 3D space, then Seedance populates the scene. Handheld camera motion and realistic shake is

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 ,

Uzay @uzpg_ 2026-06-29 22:43Z

currently the most effective way to be "data-efficient", and improve performance in a text task, given a small (eg 100-1000) dataset is to use it to select and iterate on an elicitation process that scores high for that dataset

Vonn @itsthewealth4me 2026-06-29 21:47Z

I think people associate "cost of living" without realizing how much new money was put into the markets + society, and compounded between 2019-2023 Until that is acknowledged there is no fixing certain things

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

Keith Richards by Michael Cooper, 1969

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

Dustin Tran @dustinvtran 2026-06-29 22:32Z

Believe in the length penalty. Parallel vs sequential tools, tool vs base model capability, single agent vs subagents, thinking strategy, truncation, doomlooping, exploration. You don't need new rewards, you need a better length penalty.

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

Mathias Lechner @mlech26l 2026-06-29 16:29Z

I finally managed to run GLM-5.2 fully locally by inter-connecting 256 16GB Mac Minis

GamblersAnonymous @DeepValue000 2026-06-29 21:12Z

TUNG TUNG TUNG SAHUR SET TO JOIN MILLENNIUM

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

Nathan Lambert @natolambert 2026-06-29 16:42Z

I feel like the Chinese labs I visited felt like this. It’s what Ai2 felt like. It’s the most reliable energy for making something amazing (even when you’re an underdog with the number of resources). I’m always super excited to find my next star intern.

Graham Neubig @gneubig 2026-06-29 23:31Z

We've found this sort of "sidekick" architecture to be very effective at cutting LLM spend because it allows you to do context control and not spend expensive tokens on simple tasks. Here's a 200-line example of how to do it in the OpenHands SDK :) https:// gist.github.com/neubig/412ab8d f8e6fd0b2bdf10602d77f9d86 …

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

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.

derek guy @dieworkwear 2026-06-29 23:03Z

We should first establish what the difference is between a blazer and a sport coat. A sport coat is a tailored jacket made without a pair of pants cut from the same material. A blazer is a type of sport coat, typically solid navy (but not always) and always featuring metal

Tom Schmidt ＞|＜ @tomhschmidt 2026-06-29 19:57Z

21shares updated the data .... mom get the camera!!!

Richard Seroter @rseroter 2026-06-29 20:54Z

Building up prompt debt? Beware. @dbreunig recommends the following: specify system behavior with measurements (tests) not prose stop writing prompts by hand

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

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

staysaasy @staysaasy 2026-06-29 21:59Z

Engineers: PM are never defining products enough it’s so annoying. PMs: Engineers are such babies. Fill in the gaps. It’s your job. PMs: Leadership is never defining strategy enough it’s so annoying. Leadership: PMs are such babies. Fill in the gaps. It’s your job.

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!

Ivan Burazin @ivanburazin 2026-06-29 23:04Z

There are tier 1 funds that heavily anchor on founder biography / reputation. The bet is on whether this person can hire, get deals done, and not blow up. Their downside protection is the founder's name, not the market size.

Ryan Lackey @octal 2026-06-29 17:09Z

Re the European AC Question. Our *tents* had AC.

Brandon @bran_chem 2026-06-29 20:38Z

AI moves electrons in data centers to generate new in silico designs (ideas). Drug discovery depends on moving electrons in the lab - making new chemical bonds (drugs). That’s the physical bottleneck our industry faces!

Nathan Flurry @NathanFlurry 2026-06-29 19:23Z

yep! we recently launched rivet compute as part of rivet cloud self-hosting and byoc is still core to rivet, and we built agentOS with that in mind it's significantly easier to run than sandboxes since it doesn't require nested virtualization (like microVMs) or escalated

melody kim @melodyskim 2026-06-29 23:11Z

sound of the Q train > the structure of the bridge itself played some really pretty notes > at 40s the passing train introduces a kind of “bridge” to the song > at 1:30 you can hear it decrescendo as the Q goes back underground loved this one. i’ll probably keep posting about

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

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

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!

Just Another Pod Guy @TMTLongShort 2026-06-26 14:26Z

If you understand corporate America it is fairly intuitive that for the longest time the shortest path to managerial progression was increasing headcount in your department. Every cycle you made the case for why you needed more bodies as you executed against some sort of

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.

Michael Albert, MD @MichaelAlbertMD 2026-06-29 16:49Z

My patient lost almost no weight on semaglutide. His liver healed. His blood pressure dropped. His sleep apnea lifted. His knees stopped hurting. He barely moved the scale and got better everywhere else. I'd been measuring the wrong thing for years.

Sipeed @SipeedIO 2026-06-29 17:29Z

Experimenting with different #XRay sensors and imaging techniques for the affordable mini X-ray! We've tested imaging areas of 2x2cm, 3.5x2.5cm, and 6x4cm, and the differences in clarity and resolution are huge. Which one would you go with? Vote in the following twitter!

soli @solisolsoli 2026-06-29 20:59Z

Mr. Freeze by Alex Ross

ratan @ratankaliani 2026-06-29 20:49Z

Most model evals focused on "real-world" settings (ex. TheAgentCompany, APEX-Agents, τ-bench) only target policy compliance and task success. They're useful, but they mainly treat models as tools: sandboxed settings, easy verification and primarily focused on the cost/perf

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

Aseel Swaid @aseelswaid9 2026-06-28 18:43Z

Blaze and basketball. Spanish Harlem, New York, 1975

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

47fucb4r8curb4fc8f8r4bfic8r @47fucb4r8c69323 2026-06-26 02:35Z

If we look at the percent of unhoused homeless as a percent of the US total population, we see a shocking stability across time and our current homeless population is returning to the pre-Millennial dip of the 2010s. That makes me wonder if there is a kind of min-max

ClaudeDevs @ClaudeDevs 2026-06-29 19:05Z

Boris sat down with Spotify VP of Engineering Niklas Gustavsson. Spotify ships 4,500 production deploys a day, and 73% of PRs are now AI-assisted.

USD.AI @USDai_Official 2026-06-29 17:00Z

USDAI and @0xfluid are launching a $100M DEX Liquidity Facility for sUSDai. Fluid will provide Liquidity-as-a-Service, enabling deeper markets to enter, exit, and loop with less slippage.

Architectural Art (A-A) @4AAAAart 2026-06-28 16:44Z

Soriana’ Sofa & Lounge Chairs and Ottoman Designed by Tobia & Afra Scarpa for Cassina

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

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

Alan Rockefeller @AlanRockefeller 2026-06-29 17:20Z

Mycena leana found a few minutes ago in Altoona, Pennsylvania.

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

Bohan Hou @bohanhou1998 2026-06-29 18:15Z

It's cool to learn how different warp/warpgroup roles overlap in the kernel, such as FA4. https:// tvm.apache.org/docs/tirx/nati ve_basics/cuda/profiling.html …