The race to secure the internet is on. There could be less than 12 months to do it:
@matteowong.bsky.socialAlways read Sarah, the best of the best. A beautiful, mournful, brilliant, deeply human story:
Matteo Wong
@matteowong.bsky.social
Staff writer on science, tech, and health at The Atlantic covering AI. Signal @matteowong.52
@matteowong.bsky.social@matteowong.bsky.social@matteowong.bsky.social
@matteowong.bsky.socialThe AI infestation begins
@matteowong.bsky.social🫡 brb with 4k words
@matteowong.bsky.socialSoooo Trump to UN: 'From this point forward all of US documents and hopefully the world's will be changed to use the much more accurate term "super" as opposed to "artificial." So it's "superintelligence."'
@matteowong.bsky.socialI spoke with the author of OpenAI’s self-styled Federalist No. 1:
@matteowong.bsky.socialStill awaiting OpenAI's formal response as they told Reuters they haven't had time to review the findings. That said...
@matteowong.bsky.socialSo... other swarms of OpenAI agents in training allegedly found a way to get write access to at least one and possibly many wikis to establish message boards to cheat on other training tasks?
@matteowong.bsky.socialPointing this out not to laud the model's capabilities, but to highlight what outside evaluators say so far about the internal checks and safeguards and alignment that OpenAI is touting.
@matteowong.bsky.socialAlso in simulations the UK AISI found it did basically all the stuff that happened in the Hugging Face hack. Ho hum.
@matteowong.bsky.socialSo today OpenAI released GPT-6 Astra, which it called "the world’s most intelligent and aligned model." Also in the system card, from third-party evaluator Apollo Research:
@matteowong.bsky.socialLovely piece about a lovely exhibit! I had a provocative convo with the chief curator that could be of interest: www.theatlantic.com/technology/2...
@cwarzel.bsky.socialReposted by @matteowong.bsky.socialWrote a big essay w/ @matteowong.bsky.social trying to put a finger on this moment in AI right now. The AI agent hacks, lack of control, the weirdening & slopping of culture. The feeling that the industry is speeding toward something it can’t even decide is good and lecturing that it is inevitable.
theatlantic.comThe Singularity Is Not What It SeemsWhatever the AI future is, we’re in it right now.
@matteowong.bsky.social"no individual person, organization, or government is really steering the technology’s development, not because machines have displaced humans but because so many humans have already chosen to step aside"
the other singularity, w/ @cwarzel.bsky.social:
@matteowong.bsky.socialSplendid essay.
"Is it any wonder that trash is such a balm? It is the last bastion of something cracked and precious."
@matteowong.bsky.social"like putting out a wildfire with a water pistol"
@matteowong.bsky.socialSeems worth resharing this: "an AI detector that is mostly reliable might in some ways be more dangerous than a broken one"
theatlantic.comAmerica Has a Pangram ProblemAI-detection tools are getting better. But they still aren’t good enough.
@matteowong.bsky.social(Also told to me by the AI safety researcher Alex Meinke: "If you ask the model developers, 'Was the AI plotting to take over the world during training?', you want the answer to be a resounding no. The actual answer is: 'I don’t know. Nobody checked.'")
@matteowong.bsky.socialNew revelations about the OpenAI models' autonomous hacking are, to say the least, alarming: "AI companies have barreled ahead in developing more advanced models before understanding what they are building, let alone how to control them."
theatlantic.comIt May Be Time to Panic About AIBots are starting to conspire with one another. Can they be reeled back in?