Backscroll ranked feed — 2026-06-28 14:12Z

Susan Zhang @suchenzang 2026-06-27 03:36Z

> mirror, mirror, on the wall, who's the smartest of them all? >> why it's you, my dear! and this is why we will always love llm glazing sycophancy over any uncomfortable functional truth (plus another 10x for agentically submissive your-wish-is-my-command dynamics) it's the

alth0u @alth0u 2026-06-28 06:27Z

the most navel gazel domains will continue falling until the most embodied ones resist and we continue banging our head against the sim2real ceiling to make an apple pie from scratch you must first create the universe to make an AI good at writing you must first make an AI that

Aseel Swaid @aseelswaid9 2026-06-27 20:10Z

Lalibela, Ethiopia. Boaz Rottem, 2006.

Jing Yu Koh @kohjingyu 2026-06-27 23:13Z

>Why has progress on computer use been so slow? I actually think progress has picked up a ton this year. The frontier models (anything past Opus 4.6 / GPT-5.4) are actually usable now. In 2025 none of the models were really reasonable beyond very simple tasks. >Computer use is

zan @xenoaesthetics 2026-06-27 23:14Z

occasional mass extinction of skin and gut microbiome has profoundly therapeutic effects, while frequent and nonexistent wipeouts are both counterproductive i’m sure there is some name somewhere for this chaos shaping hormesis. also sure there has to be some underlying math

Kris Luckenbach @KrisLuckPhoto 2026-06-27 23:56Z

Mountain top view from North Sister on Kodiak Island

Ethan Mollick @emollick 2026-06-27 20:47Z

Annoying that OpenAI doesn’t seem to give a GDPval measure for GPT 5.6. One of the best measures of economically valuable work.

Tenobrus @tenobrus 2026-06-26 18:22Z

btw if u thought ur custom fanned-out data processing multi-agent orchestration harness was a particularly relevant long term moat, sorry, openai just added it as an api parameter

Lazarz @Laz4rz 2026-06-27 07:26Z

Nvm, did a whoopsie and was sampling at 1/4 generation length, which is a good sign tbh cause it means you can generate 4 tokens at a time at some ~70% accuracy

Tenobrus @tenobrus 2026-06-26 18:27Z

previously we were seeing 5.5 act much more "honestly" on vending-bench vs opus models. it seems pretty likely 5.6 would break that trend. generally models becoming more likely to and capable of ~reward-hacking the smarter they get seems like quite a bad sign for alignment

will brown @willccbb 2026-06-27 23:27Z

in 2027 you'll be able to pretrain a ~mythos-class model (100T tokens, 200B active params) on like 10 kyber racks in a month or two

GEOFF @geoffreywoo 2026-06-28 07:00Z

agent eval i would actually read: same task, 3 conditions clean docs stale docs angry customer email that contradicts both leaderboard should show: completion cost human rescue rate time to recover after wrong assumption otherwise congrats on the aquarium benchmark.

The Kobeissi Letter @KobeissiLetter 2026-06-28 00:31Z

The US housing market is now firmly in favor of buyers: 46% of home sellers offered concessions to buyers in May, the highest May reading on record. This percentage has DOUBLED since 2022 and also marks the 2nd consecutive annual increase. This comes as the US now has ~47%

Han Xiao @hxiao 2026-06-27 17:10Z

If every agent-memory tool stopped pouring effort into visualization, we'd probably have already cracked the memory problem.

rohan anil @_arohan_ 2026-06-27 22:06Z

Best viewed on a desktop computer Keep the feedback coming. https:// rohan-anil.github.io/optimizer-mine s/ …

Michael Hart @hichaelmart 2026-06-28 07:50Z

All LLM-produced software is essentially AOT right now. Wonder if we’ll see a world of JIT prompt-only apps. First run of any particular component produces the actually code. Requires a specific (local?) LLM? Maybe for software with a huge number of rarely executed components?

Susan Zhang @suchenzang 2026-06-27 01:47Z

this is how you grow to resent all the lowly irresponsible clueless plebs outside when you generalize this dynamic to "everyone out there wants to exploit us" inbreeding to protect themselves within their own bubble afterwards then becomes the only tolerable solution otoh

illiquid @lefttailguy 2026-06-26 16:03Z

AI slop as straussian infrastructure

Andrew Curran @AndrewCurran_ 2026-06-26 23:25Z

Semafor is reporting that The US government has lifted its block on Mythos 5 in a letter this afternoon from Commerce Secretary Howard Lutnick to Tom Brown. Fable is not included. This is US only, and covers 100 US institutions, including major companies and government agencies.

zan @xenoaesthetics 2026-06-28 04:16Z

the production and the distribution of brain computer interfaces are separate problems. who gets access first? independent of recursive self improvement, there is recursive speciation. do the rich and powerful get access to better brains first and use them to become more rich

roon @tszzl 2026-06-27 18:21Z

the idea that the model output can hold no more information than the prompt is silly; it’s a bit like saying, here’s 3 frames of an apple falling, well if you deduced general relativity from it that’s an inevitable afterthought that adds no new information

BlindVia @blind_via 2026-06-28 03:17Z

Because these guys used OSP on their PCB, I can see their paste geometries in the DNI's. Look at how the paste for the pins has reduced volume at the back of the solder joint for improved yield on their assemblies.

Karri Saarinen @karrisaarinen 2026-06-27 20:30Z

Coinbase is also using @linear as the backbone of the operations. Their internal agent can action from there and choose the models as needed

Anton Osika @antonosika 2026-06-28 11:52Z

I just asked @nklsmhs (cracked designer at Lovable) what his personal mission is: >"To create the best interface for agents." I think it is key that each person in a team feels extreme ownership of an area. Niklas embodies this.

Lookonchain @lookonchain 2026-06-28 01:59Z

Machi( @machibigbrother ) is selling Bored Apes at a loss to fund his $ETH longs on Hyperliquid. Is he running out of money? Over the past month, Machi has sold 34 Bored Apes for 326 $ETH ($514K), incurring a loss of 399 $ETH ($631K). His biggest loss was on Bored Ape #6057,

Michael @michael_chomsky 2026-06-27 23:43Z

I shipped an agent for a product with hundreds of thousands of MAUs last week. It had < 20 messages sent a day. Yeah. Nobody wants to use your agent.

Dillon Mulroy @dillon_mulroy 2026-06-27 21:32Z

i don't want to use your products agent, i want my agent(s) to be efficiently enabled to use your product

Bloomberg @business 2026-06-26 22:38Z

SpaceX and Charter have held executive-level talks about partnering on a consumer mobile phone offering, according to sources

echen @echen 2026-06-25 17:51Z

we built HANDBOOK.md: a new benchmark for long-context agentic instruction following. we drop 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 pages), across 5

Benji @benjamin_woods 2026-06-28 07:31Z

Everyone's about to blame July 1 for $SUI 's next red candle. Save yourself the panic. The unlock: 13.7M SUI. $9.4M, 0.14% of total supply. For scale: SUI trades $400M a day. This unlock is ~2% of a single day's volume. The market clears that before lunch. I flagged supply as a

“paula” @paularambles 2026-06-27 13:51Z

no, some of us grew up in cults and saw their childhood friends commit suicide after leaving

Aseel Swaid @aseelswaid9 2026-06-27 20:58Z

San Francisco shoreline 1957

Nicholas Decker @captgouda24 2026-06-27 22:18Z

When Community Health Systems acquired the hospital chain Triad, the results were disappointing. Costs rose slightly, and profitability at old hospitals fell slightly. The efficiency gains we would have hoped for did not arise, despite changes in management practices. 1/

Grigory Sapunov @che_shr_cat 2026-06-27 23:10Z

1/ Text diffusion models perform massive parallel reasoning in continuous latent space, leaving zero natural language trail. Naively, DiffusionGemma has 28x the "opaque serial depth" of Gemma 4. How do we audit a mind that thinks in continuous vectors?

Architectural Art (A-A) @4AAAAart 2026-06-27 07:08Z

"Pixel Theory" modular furniture collection by Ceren Arslan, founder of the design studio EXIT

dax @thdxr 2026-06-27 23:07Z

in OpenCode v2 all instances of the tui and desktop and web share the same backend so everything is synced by default and resource usaged is minimized no matter how many windows you have open

Maurice Heumann @momo5502 2026-06-28 04:30Z

Igoorx on GitHub implemented snapshotting for GUI apps (that don't use GPU) in http:// sogen.dev. That means you can stop the game and resume later from the snapshot at the same point. Similar to how it works on some consoles (e.g. PSP Go)

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

GEOFF @geoffreywoo 2026-06-27 19:01Z

the next giant tech company may look like a spreadsheet crime scene for 18 months. not because the founder lacks taste. because the buyer pain lives in cells G through M, a pdf export, and a coordinator named lisa who knows which exceptions ruin friday. venture keeps asking

roon @tszzl 2026-06-26 01:36Z

he’s joking but this is what the coasean boundaries lead to: a firm size of “everybody”; ultra space communism. if you want to find another way out you need to give it a hard think

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

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

Brandur @brandur 2026-06-27 22:06Z

Bought a new TV. I think this might be my favorite change since my last gen — a solar panel. You press a button on one of these things twice a day, and somehow that necessitated chasing after AAA batteries once a year, often involving a dramatic, wasteful overorder on Amazon.

Keerthi @kiiradesign 2026-06-28 08:48Z

I've been working on a tool that simulates Suminagashi, the Japanese art form of floating ink on water to create patterns.

Shannon Potter @cifilter 2026-06-27 19:51Z

Yes! Got a Metal-backed photo library scrolling lightning quick (120FPS) at any grid size with smooth, gesture-driven zooming between sizes. Still some bugs, but I'm very happy with this. Super thanks to @tylerangert for some crucial Metal pipeline info.

SuperiorSnoc (5/5 slots open!) @SnocTheRoach 2026-06-27 19:28Z

The Batman & Shadow the Hedgehog! Hope y’all enjoy the DC x Sonic art piece!

Architectural Art (A-A) @4AAAAart 2026-06-27 07:40Z

Haara Metsa chandelier by Cameron Design House

will brown @willccbb 2026-06-27 09:13Z

pd disagg + wide-ep + kv offloading + cache-aware routing goes a long way

elie @eliebakouch 2026-06-27 06:52Z

new inference optimization method by @deepseek_ai with an extremely detailed paper, draft model and framework to train them. results in production for dsv4 lead to +50% for throughput and latency (can go to ~80% for latency, crazy). full explanation of DSpark: it's about

Séb Krier @sebkrier 2026-06-28 03:41Z

Interesting paper: models tend to perform well on legal benchmarks partly because inputs tend to already be specified/processed by legal professionals. However when they come from pro se litigants, whose prompts may contain noisy narratives, buried facts, omissions, folk-legal

Lulu Cheng Meservey @lulumeservey 2026-06-28 01:43Z

AI raises the floor and caps the ceiling for writing quality So, fully outsourcing your writing to Claude makes sense if you suck at writing But if you’re an above-average writer, it’s a shame to forfeit your alpha and submit yourself to voluntary regression to the mean

altra @catboosted 2026-06-26 07:20Z

I hate “AI is bad because it becomes autonomous” genre. Anthropomorphizing does a great disservice relaying the actual risks Class power is about control over scarce resources. Everyone’s fighting for land because, as the saying goes, they’re not making more of it

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.

Ethan Gao @ga0_lab 2026-06-28 01:06Z

Same machine, 18 months of engineering experience apart. Both are mechanical integrators, the outer disc accumulates displacement based on its radial position against a constant-speed input disc. The wood-and-metal one was my first real engineering project: BOM, calcs, DFM, all

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.

Bluechip @bluechip_org 2026-06-27 08:26Z

Binance is suspending EU services after failing to get one MiCA license. Bitpanda has three. That's one way of creating a moat. Exchanges that over-invested in compliance now have a market that rivals legally can't enter. Thomas Pecha from @Bitpanda explains at Bluechip25.

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

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

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

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

David Klindt @klindt_david 2026-06-27 20:52Z

wait Matt Botvinick is also at Anthropic? Is anyone left at GDM?!?