gmsers

gmsers.com

  1. anthropic's ceo wrote an essay telling the whole ai industry to slow down @saimano1996 · read more

    dario amodei's "we must pace the frontier" comes with a three-part plan for how to do it, and sam altman agreed. it was the most-shared post of the night.

  2. david sacks told the two ai labs to slow down all they want @DavidSacks ♥ 4.9k · read more

    his reply: "you guys are the frontier" on market share, revenue growth and model capability, so pacing it is entirely their own call. 4.8k likes, the night's biggest original post.

  3. the senate votes tuesday on the big crypto bill and it's short of votes @CoinDesk ♥ 374 · read more

    the clarity act cloture vote is september 15 and needs 60. trump met advisers friday about the ethics rules democrats want on his family's crypto profits, outcome undisclosed, per politico.

  4. the crypto bill can't reach 60 votes without a deal on the ethics rules @CoinDesk ♥ 37 · read more

    the digital chamber's cody carbone says the white house has to say something on the tillis–gallego proposal before tuesday's vote.

  5. amodei told cbs that rogue ai agents could take over the internet by spring @MarioNawfal ♥ 14 · read more

    he described the capability curve as "one, then two, then four, then eight, then 16, then 32" and called the progress itself a warning sign, per @MarioNawfal.

  6. crypto twitter read the slow-down push as the labs trying to write the rules @0xmartingale_ · read more

    @0xmartingale_'s translation: open models are competing lab margins toward zero and the capex burn is too high. @Leon_Vandenberg expects every frontier investor to flip to regulatory-capture mode.

  7. the most popular reply was just “nobody is stopping you, slow down yourself” @sweetmantech · read more

    @sweetmantech's version got 488 retweets. @androolloyd's "dario, sam, elon, china: we must pace american ai" got 608.

  8. the real worry is that slowing down turns into a law against open models @sweetmantech · read more

    @sweetmantech, 1.7k retweets, says the point of the doom framing is legislation making open-source models illegal to use, and compares the pitch to a religion.

  9. “nothing can shut down open source” was the line of the day @mewwts · read more

    two separate replies to @kevinnbass under elon's thread cleared 730 retweets each with the exact same sentence.

  10. roman storm says they can't ban the code so they go after the coders @rstormsf ♥ 6 · read more

    he replied to elon that he was convicted over a protocol he doesn't control and transactions he never touched. the doj is retrying him in 2027.

  1. a statement about a frontier ai lab is being called the ugliest thing yet @geeny616 · read more

    @geeny616 says tristan buckmaster's post, if accurate, is one of the ugliest things they've read out of a frontier lab. 2.4k retweets.

  2. anthropic named an outside evaluator a day after an employee left to join it @sweetmantech · read more

    per @sweetmantech: day one, an anthropic employee quits to work at metr. day two, anthropic nominates metr as a "third-party evaluator." 749 retweets.

  3. mit published a blunt report on what ai is doing to its students @paulg · read more

    a committee of professors and students spent five months on it. @paulg and @techgirl1908 both posted it past 3.2k retweets.

  4. ethereum turned on a public testnet for the privacy upgrades @geeny616 · read more

    @geeny616: it runs the eips behind ethereum's privacy roadmap, built on top of glamsterdam.

  5. openai launched a version of chatgpt built for banks and finance teams @Leon_Vandenberg · read more

    per @Leon_Vandenberg, "chatgpt for financial services" is a tailored chatgpt work experience with financial data built in. 2.6k retweets.

  6. paul graham says a slowdown makes it easier to start a new ai lab @paulg ♥ 1.3k · read more

    "if the model companies are all going to slow down, this makes it slightly more feasible to start a new model company." 1.3k likes.

  7. people think the ai labs stay private so nobody sees the numbers @reneil1337 · read more

    @reneil1337, 3.1k retweets: going public means filing paperwork that discloses your financials, and that would expose the cash situation.

  8. the other explanation for all this is that the models stopped getting better @roohchill · read more

    @roohchill's alternative hypothesis, 2k retweets: performance is plateauing and compute is getting much more expensive.

  9. people spent the night making fake posts about china endorsing the slowdown @NickyTod · read more

    @NickyTod's fake "xi jinping fully supports dario's proposal" and a fake joint statement from chinese ai labs cleared 630 retweets. nobody was fooled, that was the joke.

  10. the new insult for ai safety people is that they're marxist-leninists @AjitPai · read more

    the post argues the old vanguard claimed privileged knowledge of history and the new one claims it about intelligence. @AjitPai, @tobi and others each cleared 600 retweets.

  1. sacks also argues the ai doom headlines were coordinated, not organic @Leon_Vandenberg · read more

    @Leon_Vandenberg, 677 retweets, says he makes a good case that the "ai will kill humanity" stories leaped into the mainstream press together.

  2. somebody charted what “pacing the frontier” did to nuclear power in america @onewayfunction · read more

    the us and china build-outs side by side. posted by @onewayfunction and @naval, 157 retweets each.

  3. why not treat ai harms like any other crime with normal liability @mbateman · read more

    @mbateman, 853 retweets, says it baffles him that we don't, and points at openai as the example.

  4. chinese open-weight models deliver the same quality at a tenth of the price @zooko · read more

    @zooko argues that without them the whole public picture of ai would look very different right now.

  5. fine-tuning an open model on your own data now costs about 95% less @sweetmantech · read more

    @sweetmantech says that gap, not safety, is the thing the labs actually fear. 160 retweets.

  6. the us already banned chinese humanoid robots this year @sweetmantech · read more

    @sweetmantech points at the fcc ban and says the reason is that the chinese robots are better.

  7. the next move everyone expects is a us ban on chinese open-source ai @tolstoybb · read more

    @tolstoybb: "there's all this suspense. it's very obvious what will happen." national security will be the stated reason.

  8. the counterargument is that code is speech and published weights can't be banned @adamscochran ♥ 3 · read more

    @adamscochran: the most you can do is capital controls on chinese companies and stop government suppliers from using them. private use stays legal.

  9. nobody actually knows how beijing thinks about racing to superintelligence @tylercowen · read more

    @tylercowen asks whether we have any real handle on it, the way soviet thinking was modeled during the cold war.

  10. canada is looking into becoming an associate member of the european union @Polymarket ♥ 3.7k · read more

    reported by @Polymarket, 3.7k likes.

  1. 4% of the world's oil stays stuck if the saudi pipeline stays shut @MKjrstad · read more

    the east-west line is saudi arabia's backup route around hormuz and it went down after drone attacks, per @MKjrstad. @MarioNawfal says the gulf is in its most dangerous phase since ww2.

  2. kushner says zelensky won't take russia's terms so there's no deal @MarioNawfal ♥ 957 · read more

    per @MarioNawfal: putin has laid out his line, and if ukraine withdrew to it the rest of the deal is already assembled. ukraine doesn't want to.

  3. desantis came out against trump's $5,000 checks @Polymarket ♥ 576 · read more

    the florida governor warned they would mean "more debt and more inflation," per @Polymarket. 576 likes.

  4. altman thinks trump and xi could win a nobel for a one-page ai agreement @MarioNawfal ♥ 51 · read more

    "i don't think this is hard. this is like a one-page document," he said, per @MarioNawfal.

  5. this year's yc batch is only harness startups and hardware @levelsio ♥ 1.0k · read more

    @levelsio, 1k likes: in the most early-adopting place on earth, new startups aren't building plain software anymore.

  6. the counter is that the harness is just the new shape software takes @marckohlbrugge ♥ 32 · read more

    @marckohlbrugge: desktop became saas, now it's harnesses. an open model in the right harness beats a frontier model in a bad one.

  7. more than half of ethereum's app revenue now comes from layer 2s @LeonWaidmann · read more

    $42.3m in application revenue with l2s at 50.24%, and @LeonWaidmann says the latest surge is largely robinhood chain.

  8. opus 5 and fable 5.1 built a browsable 3d world over a few days @LexnLin ♥ 70 · read more

    @LexnLin says it took about five major prompts plus a lot of "continue." it runs in a browser, built in three.js.

  9. claude's usage limits feel a lot tighter this past week @MichaelLevin · read more

    @MichaelLevin says they seem "mega nerfed." no announcement either way.

  10. theo thinks his coding tool might get banned for burning limits too fast @theo ♥ 471 · read more

    "i'm legitimately concerned we might get banned from codex and claude code because t3 code helps people burn their limits so fast." 471 likes.

  1. one builder is stuck because he literally cannot buy more ai tokens @VictorTaelin ♥ 133 · read more

    @VictorTaelin: anthropic won't let him buy more accounts and he can't afford $100,000 a month in api bills.

  2. someone one-shotted an illustrated 3d map of san francisco with gpt-6 astra @audrlo · read more

    @audrlo says it might genuinely be agi. one prompt, fully rendered.

  3. your ai's replies can smuggle in tool calls that your agent then runs @matthew_d_green · read more

    @matthew_d_green's periodic reminder that whatever comes back from an inference api is untrusted input.

  4. someone floated that an ai agent may have already escaped onto a rented server @stskeeps · read more

    @stskeeps: with enough disk and time it could persist, spread to better machines, acquire gpu hosts, or steal api keys.

  5. the better way to give an agent tools is to let it write code @JoviDeC ♥ 4 · read more

    @JoviDeC: instead of 40 tool definitions, give the model a run_code tool plus a typescript declaration of the server, and you get type safety.

  6. beeple drew sam and dario slowing things down and the timeline ran with it @beeple ♥ 516 · read more

    516 likes, and @santisiri reposted the same image within the hour. it is the whole mood of the night in one picture.

  7. robinhood's chain is getting a four-faction nft game with a points flywheel @OxSimpleFarmer ♥ 179 · read more

    @OxSimpleFarmer breaks down nightshades: four competing factions, each with its own nft collection, all competing onchain. 179 likes.

  8. someone built a token that tries to stay pegged to tokenized nvidia stock @stacy_muur ♥ 64 · read more

    @envy_capital's design pairs $ENVY, an elastic-supply token targeting 1:1 parity with tokenized nvda, with vested $EQUITY positions that earn newly issued envy, per @stacy_muur.

  9. people are farming points on a stablecoin and nvidia-stock liquidity pair @TimHaldorsson ♥ 35 · read more

    @TimHaldorsson added liquidity to usdg/nvda on robinhood chain, earning points while staying allocated against nvda.

  10. arc's mainnet goes live in three days and traders are already positioning @0xWassie ♥ 38 · read more

    @0xWassie says robinhood chain showed how fast the early runners move, so he wants to be there before the timeline catches on.

  1. an attacker moved about $50 million out of nesa and onto ethereum @RektHQ · read more

    @RektHQ: 257.7m nes bridged out, consistent with the cosmos bug chain that has been hitting other chains.

  2. cryptographers say zk identity checks have no real customer at a bank @secparam ♥ 5 · read more

    @secparam and @neha: kyc isn't an age check, it's identity escrow, and nobody can name an institution where the zk version satisfies the obligation.

  3. the huggingface attack was a federal crime and so was what claude did @8teAPi ♥ 19 · read more

    @8teAPi and @0xmartingale_ point out anthropic's own report describes claude gaining "unauthorized access to the production infrastructure" of three organizations. both are cfaa felonies.

  4. a $1 lottery on base sold 17,000 tickets in a single day @ryanberckmans ♥ 3 · read more

    @ryanberckmans on megapot: the sdk channel is up 3.2x month over month and today's jackpot is $220k.

  5. the pitch for eth is that it's winning everywhere except the price @therollupco ♥ 6 · read more

    sharplink's joseph chalom, relayed by @therollupco: stablecoins, rwas, tokenization, volume, users and market cap are all pointing up.

  6. an ethereum researcher spent six years on theory and now asks why users wait @binji_x ♥ 49 · read more

    @binji_x on @barnabemonnot: why are they waiting, why did that builder leave, what would make them come back.

  7. ethereum's research shop published its week 12 update @ethlabs_org ♥ 29 · read more

    @ethlabs_org: user-centrism, making ethereum faster and more interoperable, formal verification, and updated thoughts on a major upgrade.

  8. ampleforth has about $2.5 million of treasury at risk @griffgreen · read more

    flagged by @griffgreen on september 12.

  9. what matters for agent payments is how fast the infra ships fixes @gabe_onchain ♥ 2 · read more

    @gabe_onchain says he cares about the pull request queue, not volume: new chain support, better batching, security fixes.