OpenAI 推日常任务基准 LifeVal

For AGI to benefit all of humanity, it should help people with the everyday tasks that matter most to them. So we built LifeVal, a new benchmark shaped by feedback from thousands of ChatGPT users on where they most want help from a personal assistant—from finding childcare to accessing government services to managing personal finances. Over the past year, we’ve used LifeVal to measure and improve how well our models handle these everyday tasks. https://t.co/GlUjixzEM0

ChatGPT and Codex are helping engineers build for the physical world. We’re sharing our progress across mechanical design, electrical engineering, and simulation—from helping our hardware teams iterate faster to testing new workflows with industry collaborators. https://t.co/b3rW9zYJGA

As AI becomes more capable at finding vulnerabilities, defenders need to put it to work before attackers do. We’re planning our next internal security sprint at OpenAI and partnering with Cathedral Systems, whose researchers will use our models alongside our teams to challenge our defenses. They’ll bring differentiated expertise and approaches to our ongoing work to find and fix weaknesses with AI. We’re also raising bug bounty rewards, including more than quadrupling rewards for critical findings. Exceptional findings will be eligible for up to $172,000, including bonuses. We’ll keep advancing our models, using them to strengthen our own systems, and expanding access for trusted defenders. https://t.co/vcaoHROTt2

RT @SkinnerPm: A truly disgraceful public official. A blatant racist and dishonest appeal

RT @haider1: Yann LeCun says: People in AI have made the same mistake for 70 years: if a machine can do a task humans see as intellectuall…

RT @JeffDean: Election Tuesday is just two days away. For those eligible to vote in the U.S., please make time to go to the polls to vote…

RT @lukaskuhn77: Introducing LeVJEPA 1.5 Large, the first step in scaling our method. It closes the gap to V-JEPA on both image and video…

RT @KenRoth: The Trump administration is openly meddling in European democracies to promote far-right political parties that support Putin…

RT @randall_balestr: Another example that we should spend more time thinking about improving pretraining losses rather than just searching…

Nothing will ever be the same for developers. It's just incredible that you can build a complete application and deploy it in under 5 minutes. @vercel added Python support, and I'm in heaven! https://t.co/ZVDo8lVmTu

This is a new developer platform for fine-tuning models. You provide: • The tasks the model should solve • The data the model needs to solve them • The criteria to evaluate the results @trajectorylabs will take this and handle post-training, evaluation, and deployment.

RT @tonyzzhao: This is a great model. Congratulations @DrJimFan and @yunzeman for the awesome work!

This is an open-source tool that will let you carry knowledge between different coding assistants. This is super cool for moving between assistants when you hit a usage limit. Basically, this will record your coding sessions and turn useful fixes, corrections, and workflows into lessons. These lessons will then become available to other coding assistants.

This is a new decision model for audio input: • You provide the audio • You provide a list of questions • Model outputs answers + confidence scores The model analyzes the audio directly (no transcript needed). Something really cool is the model's ability to answer questions based on tone, hesitation, laughter, and other audio clues that wouldn't be in a transcript.

This is a new streaming decision model: • Give it text + list of questions • It answers back with a confidence score • Answers update as the text arrives, 3x faster But it's also an extraction model: • Text in, structured data out • ~150 ms per extraction • Claims 75x cheaper than Opus 5.5 on document extraction It all uses the same endpoint. They are offering 20M free tokens to start.

RT @GreylockVC: One of the biggest bets startup employees make is on the founder. That trust may not be tested until years later, when the…

RT @svpino: We haven’t figured out how to scale from producing stuff to shipping stuff: • 2,450% increase in lines of code • 240% increase…

We haven’t figured out how to scale from producing stuff to shipping stuff: • 2,450% increase in lines of code • 240% increase in commits • but only 30% increase in releases This video is really cool: https://t.co/Oi1reBDNL8 https://t.co/8Xs0GyFJ7n

I'm adding every account that replies with AI-generated slop to a list, so I can mass-mute and block them all. This is probably an uphill battle, but I'll try to fight back because the replies on this platform are useless right now. I'm using Grok to identify slop, and based on my tests, it does a decent job. To avoid false positives, I'm only blocking accounts that consistently show up as bots. Even if a human reply gets caught up once or twice, the bot is unlikely to classify it incorrectly consistently. I'll share the list later, so you can ask your bot to mute everyone there as well.

Specialized Models >> Larger Models. This is an open-source platform that will help you choose and run specialized models in your applications. Basically, you can set up which models you want to serve specific tasks. For example, if your agent needs to classify a message, read text on an image, search files, and write a response, you could have 4 different models to serve each of these tasks.