研究人员正在探索AI的新前沿，从优化直接偏好优化（DPO）的超参数景观，到使用随机非线性增强视觉识别模型中的不确定性估计。与此同时，OpenAI已正式推出其Decisions API，提供更快、更具成本效益的有界答案，而Ethan Mollick发布了一本新书《共存》（Co-Existence），探讨人机协作。讨论还涉及AI泡沫破裂的可能性以及AI在网络安全中的作用，并发现了PHP中的古老漏洞。 AI
影响 DPO优化和不确定性估计的进展可能提高模型的性能和可靠性，而OpenAI的Decisions API旨在加速应用程序开发。
排序理由 该集群包含多篇关于AI研究和开发的学术论文和讨论，而不是单一的主要产品发布或行业事件。
AI 生成摘要 · Google Gemini · 来自 20 个来源。 我们如何撰写摘要 →
完整方法见我们的编辑标准。
arXiv:2610.27895v1 Announce Type: cross Abstract: Direct preference optimisation (DPO) is widely used for language model post-training, but its performance depends strongly on the choice of its hyperparameters, such as the learning rate and the reference-policy regularisation par…
arXiv:2610.29875v1 Announce Type: new Abstract: Reliable uncertainty estimation is important across visual recognition tasks, providing information about the confidence and reliability of model predictions. Many uncertainty estimation methods obtain predictive diversity through m…
My new book, Co-Existence, is out today. It is a guide to how to live and work in a world where AI is sometimes, but not always, better than you. (And, for the release, I worked with Opus 5.5 to make a quite interesting experience for the book's new website: co-existence.ai)
<blockquote> <p><strong>Attributed compile + one small offline smoke test</strong><br /><br /> <strong>Primary sources (OpenAI, 2026-10-22):</strong> <a href="https://openai.com/index/decisions-api" rel="noopener noreferrer">Build more responsive apps with the Decisions API</a> (…
Nu #PRO op 1 staat, gaat rechts uit zn dak over het gebruik van #AI . Een AI-cancel is een mooi resultaat 💚❤️ #politiek RE: https://bsky.app/profile/did:plc:xpftwxnjd6qa4c72i2grsch5/post/3myef3alef52f
Every honest account of running an agent eventually has to say what it does not hand over. Mine is about a fifth of the work, and the reasons cluster into four kinds. 1. Anything where being wrong is expensive and silent The agent is good at tasks where a mistake shows up immedia…
“Warzel: I am amazed by how much the infrastructure in certain areas of the internet can hinge on one or two people. There is a fragility to all of this that is concerning. So what do you want to see? Surý: When I started doing open-source, it was fun. And it kind of still is. Bu…
Fear mongering by ai ceos is laughable. They create a service, claim it is dangerous, and keep working on it for fear that the fears of the masses dissipate and people will realize... Who smelled it dealt it. That can't be good for share prices. # ai # business
At this point it's very likely that the AI bubble burst, if it happens, will be presided over by a Democratic-majority Congress. I hope # NoKings is prepared to stand up to their own party when the phrase "too big to fail" starts getting thrown around. # uspol # election2026 # ai
PHP armageddon: OpenAI, Google’s Wiz find ancient holes, panicking the internet This is the kind of security issue that was found and prevented by AI, but is also a high risk because of AI. https:// techcrunch.com/2026/10/21/php- armageddon-openai-googles-wiz-find-ancient-holes-p…
🤖 PHP Armageddon: OpenAI, Google’s Wiz find ancient holes, panicking the internet This is the kind of security issue that was found and prevented by AI, but is also a high risk because of AI. 📰 Source: AI News & Artificial Intelligence | TechCrunch 🔗 Link: https://techcrunch.com/…
The primary purpose of fast food is not to make people less hungry, it is to be sold. This is not a post about food. # AI
🦑 Translation is not a button you press. It is a conversation between languages, cultures, and people. AI can produce words in another language. That does not mean it understands what those words are doing there. It doesn't know when a joke is supposed to sting. When a double mea…
Pondering AI - and AIthics A blog by Justus There is no shortage of discussion about AI, and LLMs in particular. This is my personal take on the debate: the claims I question, the concerns I share, and the questions I am still working through. Wild claims abound, from “Soon, we w…
Nu #PRO op 1 staat, gaat rechts uit zn dak over het gebruik van #AI . Een AI-cancel is een mooi resultaat 💚❤️ #politiek RE: https://bsky.app/profile/did:plc:xpftwxnjd6qa4c72i2grsch5/post/3myef3alef52f
What Lawmakers Know About Data Centers That You Don’t Tbh, it is stuff like this that has many times made me feel embarrassed to be part of the pivotal moment in tech. There's a better way of doing this. This is just pure, unadulterated corruption. # ai # llm # datacenters https:…
I used to do a bit of progamming so this is about AI. It seems to me that AI is like a pretty accomplished compilator, you know, an Integrated Development Environment (IDE) like we used to call them, but hey those IDE's are a bit more avanced today obviously. However, calling the…
the LLM doesn’t know anything. it’s not taught, it’s trained. as i understand it, it spits out the sequence of words that it calculates to have the highest probability of being “correct” based on all the words that have been fed to it thus far, along with their respective credibi…
RE: https:// tldr.nettime.org/@tante/117472 909371855161 It is really hard to accept that this actually is Jensen Huang and not a deep fake. Good thing those "digital factories" can all be powered by "clean" energy in Germany /s # timeline # ai # llm # climatechange # idiocracy
RE: https:// wandering.shop/@aesthr/1174700 82684445346 This is _fantastic_. I wanted to include a quote that grabbed me, but the entire article is a rainbow of highlights for me now. So I’ll stick with something short, and something that disturbs me on the regular: "We had the m…