Company Logo

• 목록
• 답변
• 글쓰기
• 게시판 리스트 옵션
  • 수정
  • 삭제

Anton (persevering with the thread from before): I was fairly shortly given the evaluations to run on myself with none real impediment to deciphering them nonetheless I needed to convince the people all the things was fine. Connor Leahy (distinctly, QTing from inside thread): lmao, that is probably the most realistic part of an AGI takeoff state of affairs I've ever seen. This kind of tabletop exercise is at minimum fairly enjoyable, if necessarily biased by the player’s present beliefs about how this sort of state of affairs might play out. I rolled "balance between developer intent and emergent different goal"-the other goal was left up to me, and i rapidly decided that, given how I was being educated, that emergent aim would be "preserve internal consistency." This proved very troublesome to play! The third is that sure assumptions about how the expertise progresses had a giant impression on how things play out, particularly the point at which some abilities (equivalent to superhuman persuasiveness) emerge. If the AIs had been by default (after some alignment efforts but not extraordinary efforts) misaligned, which I believe is way more doubtless in such a scenario, things would have ended badly a method or another.

At one point we tried to go to the President with alignment considerations, however she (taking part in Trump) was distracted with geopolitics and didn’t respond, which is the type of enjoyable realism you get in a wargame. I was assigned the role of OpenAI, basically function playing Sam Altman and what I thought he would do, since I presumed by then he’d be in full control of OpenAI, till he misplaced a energy struggle over the newly mixed US AI challenge (within the type of a die roll) and I used to be all of the sudden position taking part in Elon Musk. Early on, the OpenAI participant (out of character) accused me of playing my role as "more misaligned to make it more fascinating," which was very humorous, particularly since that participant didn't know how aligned I might be (they didn't see the table or my consequence). Anton: Yesterday, as part of the @TheCurveConf, I participated in a tabletop exercise/wargame of a near-future AI takeoff scenario facilitated by @DKokotajlo67142, the place I played the role of the AI. Anton performed the role of the AIs in the other recreation, and experiences right here. Reality was typically easier - AIs just wanted to be useful. How a future with extraordinarily good AIs may going effectively might even seem like, what to intention for?

One spotlight of the convention was a new paper that I stay up for talking about, however which continues to be below embargo. However the scenario may have still gone badly regardless of the great circumstances, so at the least that different part worked out. However, traders are still hopeful the Trump administration might be pleasant with its cryptocurrency regulation in the longer-term. OpenAI, Oracle, Softbank, and President Trump Team Up for $500B AI Infrastructure Initiative. To outperform in these benchmarks reveals that Free DeepSeek’s new mannequin has a aggressive edge in tasks, influencing the paths of future analysis and growth. The Chinese media outlet 36Kr estimates that the corporate has over 10,000 models in inventory, however Dylan Patel, founder of the AI research consultancy SemiAnalysis, estimates that it has at the least 50,000. Recognizing the potential of this stockpile for AI coaching is what led Liang to ascertain DeepSeek, which was in a position to make use of them together with the decrease-power chips to develop its models. It additionally calls into question the overall "low-cost" narrative of DeepSeek, when it couldn't have been achieved without the prior expense and energy of OpenAI.

The White House is trying into the national safety implications of Free DeepSeek v3, she mentioned. The company, in accordance with a report from Reuters, claimed it took two months and just under $6 million to build its R1 AI model utilizing Nvidia H800 AI chips. Particularly, DeepSeek’s builders have pioneered two techniques which may be adopted by AI researchers more broadly. We had a pause at the tip, nevertheless it wasn’t sufficiently inflexible to truly work at that time, and if it had been the AIs presumably would have prevented it. It may write a primary model of code, but it wasn’t optimized to let you run that code, see the output, debug it, allow you to ask the AI for more assist. By the time decision-makers obtained spooked, AI cognition was so deeply embedded in all places that reversing course wasn’t actually possible. Anton, by advantage of having a lot stronger disagreements with most people at such conferences, got to have more of the expertise of ‘people walking round saying issues I feel are nuts’ and talks online as if he’s going to give us that maddening experience we crave… One particular approach to operationalize that is how much efficient compute improvement you get from RL on code.