Daily Journal
Avatar de Kevin Flyangolts

Kevin Flyangolts

@kevfly16

@aclid

25 postsÚltimo em 13 de set, 19:58Atualiza a cada 15 min

  1. ver no X

    @DavidRBellamy @DavidSacks Is this specifically about rogue AI or includes people trying to misuse AI to accomplish this?

  2. ver no X

    @paulg It’s also surprising how few people will actually try to duplicate your idea and work on it for long enough to matter

  3. ver no X

    @Noahpinion 💯

  4. ver no X

    @alexiskold Cloudflare

  5. ver no X

    Frontier AI agents can write production pathogen surveillance pipelines when handed a detailed spec. When handed just the raw sequencing data and a minimal prompt, they fell short. The first benchmark measuring agent performance on pathogen surveillance: benchmarks.bio/surveillance

  6. ver no X

    @kenbwork Maybe there are still new end goals to be discovered in biotech

  7. ver no X

    @DNA2FUNCTION @ATinyGreenCell Those aren’t the only two options. If AI is smart enough to make novel seqs, it’s smart enough to detect them too. An ambiguous seq that may be harmful in theory is going to be as challenging to detect accurately as it would be for a nefarious actor to misuse it

  8. ver no X

    @DNA2FUNCTION @ATinyGreenCell There are no plans to reject ambiguous or novel sequences that can’t be linked to a harmful function

  9. ver no X

    @DNA2FUNCTION @ATinyGreenCell DNA orders are extremely rarely outright rejected. 99% of the time even orders that are flagged resolve without affecting shipment times. What gets blocked are flagged orders without appropriate justification and documentation that would be regularly expected for the research

  10. ver no X

    @DNA2FUNCTION @ATinyGreenCell DNA costs are by and large not the major costs of biotech startups. Equipment, lab space, labor, clinical trials, and regulatory approvals are 10x to 100x more expensive and more cost prohibitive than DNA

  11. ver no X

    @BioAndBaseball @001TMF Many of the screening practices called out in the letter are already in place and have been for the last decade at many of the largest synthesis providers. From what I’ve seen R&D has only gotten faster with gene length constructs approaching oligo level delivery time

  12. ver no X

    @DNA2FUNCTION @ATinyGreenCell @OpenAI @AnthropicAI @idtdna @TwistBioscience @thermofisher As with any dual-use product, we need to strike the right balance. We are making every effort to not get in the way of legitimate research. The KYC process are being designed with these very issues top of mind

  13. ver no X

    @agupta @owl_posting If a bioterrorist is trying to make a bioweapon the scenario is far more balanced and the approach as is could be effective. For state actors, preventing proliferation of all WMDs is a major. Detection is an important part of the infrastructure

  14. ver no X

    @agupta @owl_posting What scenario do you have in mind where there’s unequal access? State actors?

  15. ver no X

    @agupta @owl_posting If the models are smart enough to harness hypothesized biology, they would also be smart enough to detect when the harnessed biology will lead to harm

  16. ver no X

    @agupta @owl_posting Maybe with some more creative ideas of how to cause harm, this can already be broken. Maybe new tech is months away from breaking it. Today, it’s not trivial to circumvent and the problem has sensible solutions. I’m gonna keep building to so it’s gonna do shit 🙃

  17. ver no X

    @agupta @owl_posting State of the art screening practices are in line with current genetic engineering capabilities and AI-enabled biological design tools (e.g., biorxiv.org/content/biorxi… and science.org/doi/10.1126/sc…)

  18. ver no X

    @ATinyGreenCell @OpenAI @AnthropicAI @idtdna @TwistBioscience @thermofisher Some places to get involved / people to chat with below! Join the SBRC if you’re interested in contributing to what we screen against sbrc.bio fourtheon.bio/#about securebio.org/ai/ sentinelbio.org/people/

  19. ver no X

    @ATinyGreenCell @OpenAI @AnthropicAI @idtdna @TwistBioscience @thermofisher Please join in on the red teaming to help us make it better 🙃

  20. ver no X

    @ATinyGreenCell @OpenAI @AnthropicAI @idtdna @TwistBioscience @thermofisher We’re already screening well beyond explicit homologies doi.org/10.1126/scienc…

  21. ver no X

    @ATinyGreenCell @OpenAI @AnthropicAI 99% of researchers working in pharma, ag, industrial, and consumer will never see a difference in their work. If you’ve ordered from @idtdna @TwistBioscience or @thermofisher among many others you’ve already experienced the frictionless process I’m describing

  22. ver no X

    💯 x.com/IanRountree/st…

  23. ver no X

    We're kicking off SynBioBeta with a night of apps and drinks with our friends at @IBBIS_bio on May 5th from 6:30 pm to 8:30 pm. Register today! lu.ma/6fhjkqin

  24. ver no X

    Honored to be included in this year's class! x.com/BSPH_CHS/statu…

  25. ver no X

    Fun chat with @nlarusstone! Thanks for hosting us x.com/bitsinbio/stat…

Posts exibidos como publicados originalmente em @kevfly16. Links encurtados e mídia ficam disponíveis ao abrir o post no X.