An internal version of our next major model produced 10 new results on long-standing open problems in mathematics and theoretical computer science, using roughly $2,000 worth of tokens at GPT-5.6 Sol API rates.
Reminder to go touch grass
A scientist in Japan demonstrated that the air inside a forest strengthens your immune system for a full month
The study is incredibly fascinating:
Qing Li, an immunologist at Nippon Medical School in Tokyo, took 12 men on a 3-day forest trip. These were slow walks with no breathing exercises and no set route.
Then he tested their blood and found a surge in natural killer cells -- the immune cells your body uses to fight cancer and virus-infected cells.
Remarkably, the effect didn't fade when they went home, it was still measurable 30 days later.
A control group that walked the same distance through a city showed no change at all.
Pine, cedar, and cypress trees release compounds called phytoncides to defend themselves from insects and bacteria. You breathe them in, and Li believes they switch your immune cells into high gear.
He estimates one forest walk a month may be enough to keep those cells elevated all year.
The best medicine might not even need a prescription -- just a walk among the trees.
GPT-Live can listen while it speaks.
To make that feel natural at ChatGPT scale, we rebuilt the voice stack from client to model.
This new architecture keeps audio flowing continuously, so deeper reasoning and tool use don't interrupt the conversation.
Yes, the rumors are true: The more powerful Notebook experience is now 100% rolled out to ALL Pro users.
You might be asking... How exactly do I use it? Here are a few queries to try (ofc dependent on your sources).
Hot Tip: bookmark this post! Free users too 😉
For Learning:
"Create 5 quizzes of ascending difficulty focusing on a different topic from my sources"
"Turn my raw lecture notes into a formatted PDF study guide complete with a 5-question practice quiz."
"Add all of the URLs from this class syllabus as sources"
For Business:
"Calculate revenue growth across each product line and plot the trends"
"Read our marketing brief, search the web for competitors, and recommend how we should tune our messaging"
For Personal:
"I uploaded a bunch of receipts from our kitchen renovation, make a spreadsheet to track all the work and costs"
"Help me analyze this property purchase. Look up recent market performance in the neighborhood, predict trends, and make a spreadsheet to model all the data you find"
Firecrawl@firecrawl · 公司官方 · 5 天前
Today we're launching anydoc, our new Rust-based doc parsing engine 🔥
sub ~5ms markdown parsing for PDFs, Word docs, slide decks, and 10 other formats with best in class quality.
100% open source and powering our /parse endpoint!
引用 Nicolas Camara @nickscamara_introducing anydoc now your agents get 100x faster local parsing for pdf, docx, pptx & 10 more formats - sub-5ms md conversion - 500 docx files in 1.7s - top quality across all 13 formats - rust-based - open source already powering @firecrawl /parse github.com/firecrawl/anydoc查看被引原帖 ↗
Small Gemini API update: you can now use the Google Maps and Google Search tool at the same time with Gemini 3.5 Flash and 3.6 Flash
Been on the backlog for a long time, glad we landed it!!
Audio moves through a dedicated fast path, while deeper reasoning and tool use happen asynchronously.
We also reduced voice-session startup from six network round trips to one.
We’re releasing the manuscripts, formal Lean certificates, and reasoning walkthroughs so mathematicians can examine these results and build on their ideas.
openai.com/index/ten-advance…
Look closer 👀
The tiniest details make the biggest difference.
The results span sphere packing, coding theory, group theory, quantum complexity, lattice cryptography, extremal combinatorics, and more.
Among them: establishing the existence of non-sofic groups and exponential improvements to bounds on high dimensional sphere packing.
NVIDIA@nvidia · 公司官方 · 6 天前
Meet Alpamayo 2 Super, now commercially available for robotaxis and autonomous vehicles.
Built for complex real-world driving, this open reasoning model adds 360° awareness, high-level driving decisions and automated reasoning labels.
We built it, so you can build on it.
What are you cooking up with Gemini Omni? Drop your creations in the replies. 👇
Reminder: Free users can create up to 10 videos *at no cost* with Gemini Omni until August 4th, 2026 at 11:59pm PT.
The result is a faster, more natural conversation with ChatGPT Voice from the moment a session starts.
How we built it:
openai.com/index/continuous-…
Mathematics helps us understand how traffic moves, how diseases spread, and how cells build proteins. It also underpins everyday technologies, from GPS and weather forecasting to medical imaging.
Progress in foundational mathematics can ripple across science and technology. These advances suggest AI can help researchers push further on problems that have resisted progress for decades.
v0@v0 · 公司官方 · 6 天前
Introducing the new v0 API.
Programmatic access to v0's app-building capabilities:
• Start a chat from a prompt, repo, or ZIP
• Render a dev server preview
• Send follow-up messages
• Deploy to Vercel
Sign up today for a new Max plan and get 7 days of Unlimited Seedance 2.5 on launch.
Coming soon to Runway.
我开源了几个 skill 后发现开源 skill 获得的收益巨大，因为大多数用户不知道效果这么好是大模型的功劳还是 skill 本身的功劳，他们会把所有的功劳给这个 skill 和作者，甚至可能唯一起作用的是 skill markdown 里的某一句话，甚至 skill markdown 里某些内容会有副作用用户都感知不出来，用户只会直夸牛逼，而且随着大模型的不停进化夸牛逼的声音会越来越大，就是躺赢啊！
这就是无法量化结果的好处啊😬
引用 AlexZ 🦀 @blackanger是我太超前了吗 去年写的 Rust Skill，感觉现在随着大模型发展，才变得好用了 。。。这半年我可啥都没改啊查看被引原帖 ↗
This is insane.
If you want to learn how to build games like this, I made a guide:
somethingbig.ai/gauntlet-loo…
引用 Rishi @0xRishiIntroducing Modern Claudefare Built fully with Opus 5 on High Mode over a few days using principles from @mattshumer_ ’s Gauntlet Loop - 84,100 lines of code Includes remakes of 4 beloved maps: 0:00 - Rust 0:40 - Highrise 0:57 - Nuketown 1:26 - Terminal Solo & multiplayer (w/ voice chat), bots included Tools used: - @threejs as core Javascript 3D library (all 20 guns and 4 maps generated by the model in code) - @MeshyAI for 3D models + rigging + animations of 8 soldiers, 1 helicopter (highrise) and 1 bus (nuketown) - @ElevenLabs for 430+ SFX, Music & TTS generations Play it now at modernclaudefare.com (More details in thread)查看被引原帖 ↗
Also did a more detailed video breakdown on this across short form platforms for those who prefer watching:
On AI taking jobs: A farmer from 50 years ago would say your job today isn't real work.
When I interviewed Sam Altman a few months back, this was how he explained the future of work:
"Not only would they not believe you that this thing was gonna happen, they very likely would look at what you do or I do and say, that's not real work."
"If you're farming... you're keeping people alive. And you people of the future -- life just got too easy for you... you're playing a game to fill your time and your need to feel important."
"To us, though, it feels real. Certainly, to me it feels real."
"If we could see those jobs of the future, we would be like: okay, maybe our jobs were not as real as a farmer's job, but it's a lot more real than this game you're playing to entertain yourself."
"I have some short-term worries as we go through this transition, but I'm so willing to bet on human drives being what they are. I think we'll find plenty of things to do."
Link to the study:
pmc.ncbi.nlm.nih.gov/article…
NVIDIA@nvidia · 公司官方 · 6 天前
Read the blog →
nvda.ws/4pRK2bU
SwiftUI 必出布局问题，需要不停地 tweak
俺上电视啦! 聊聊 kimi-k3 这么猛为什么还要开源
上周有幸接受了深圳卫视-科创最前沿栏目的采访, 主要聊了聊 Kimi-K3 这次取得如此牛的成绩背后的意义. 以及SOTA级别的模型就这样开放权重了, 不怕别人抄吗?
在我来看是真不怕的, 想象一下一个武林高手, 把自己的秘籍写成书, 免费送. 那是不是就意味着这些武林秘籍不是谁都练的.
大模型的情况也是这样的, 虽然权重也开源了, 训练技术也开源了, 但是, 最重要的, 训练使用的数据和经验, 这个是没开源的.
举个最典型的例子, 现在个人拿到kimi-k3的模型权重, 然后自己租算力微调, 大概率是会把模型越调越拉跨的. 就是因为既没有优质的后训练数据, 也没有优秀的后训练能力.
而现在大模型训练的数据其实大部分都是AI合成的了. 护城河已经从“静态的代码和模型权重”转移到了“持续的进化能力和数据飞轮”了. 所以即使全面复刻了 kimi-k3 的技术架构, 但是没有高质量合成数据流和持续的后训练体系, 也是没有办法竞争的.
但是 kimi-k3 开源获得的收益却是事实存在的, 不但能获得超高的曝光, 而且还能提升自己的社区影响力, 并且还能收获社区的正向反馈. 开源出来的训练工具链, 社区优秀的工程师在使用中也会不断的帮助反馈和提升. 形成整个生态的进步. 得道者多助这个道理是不会变的.
#kimik3
#深圳卫视
#科创最前沿
The full interview:
youtube.com/watch?v=zwnVUiwO…
九原客@9hills · 中文博主 · 6 天前
哎，怀念 opus4.6 的语言能力。
引用 Zihan "Zenus" Wang @wzenus让 Claude Code 说人话的 instruction 1. 使用标准术语。已有通用说法的概念不要另造表述。 2. 不要使用比喻。如果一个说法需要读者推断其指代对象，改为直接说明。 3. 表头和分类名使用中性名词，如“问题”“现象”“影响”“结果”。 4. 不使用“是 X，不是 Y”的对比句式。 5. 表格条目不要求每条都包含数字或结论。部分条目可以只说明发生了什么。 6. 未查明原因的问题写“原因未查明”。 7. 不使用口语词。 8. 不使用拟人表述。 规则 1：使用标准术语 超长度 / 超不超长度 → 截断 / 是否触及长度上限 Ray 的端口会跟自己撞 → 多个任务的 Ray 抢占同一端口 臂 → 方案 只看长度那条线 → 长度启发式基线 刷奖励 → 奖励被优化但评测指标未改善 存档 → checkpoint 规则 2：不发明比喻作为术语 血统 → 基座来源 8 个不同血统 → 8 个不同基座的模型 那是尺寸差异，不是血统差异 → 原有区间由 Qwen 的 14B、32B、72B 构成，差异来自参数量而非基座 已经吃掉一半空间 → 已覆盖一半区间 变成了一个认题目的检索器 → 退化为问题识别，与学习价值无关 挑出来的题目看上去健康得多 → 选中问题的截断比例更低 误差范围还盖着基准线 → 置信区间与基线重叠 规则 3：表头和分类名使用中性名词 坑 / 代价和教训 → 问题 / 影响 把握 → 确认程度 怎么做的 → 实验设置 重挑一次还剩多少重合 → 重采样后的重合率 有几种不同取值 → 取值数量 还在跑的 → 进行中 一个必须关掉的开关 → 需要关闭的过滤器 章节状态标签需要保持一致。 如果前面使用： 已完成 已确认 未达到基线 结论待定 后面出现了： 做通了 意外发现 需要修 没有买到想要的 统一使用前面部分的中性状态标签。 规则 4：不使用“是 X，不是 Y”的对比句式 梯度对齐：是噪声，不是信号 → 梯度对齐：三项检查结果均在噪声范围内 我们买到了血统上的多样性，但没有买到能力上的多样性 → 新增模型覆盖了更多基座，但正确率均低于原有区间下界 考法和用法对不上 → 评测口径为成对比较，与实际使用方式不一致 稳定是靠粗糙换来的 → 取值数量少的指标重合率高 越稳定的指标越挑不出东西，越挑得出东西的指标越不稳定 → 重合率与取值数量呈反向关系 规则 5：允许条目不包含数字或结论 例如： 对照组为随机选取的 32 个问题。 这条内容已经完整，不需要继续补充其他方案与对照组的关系。 规则 6：未查明原因时直接写“原因未查明” 前文已经写明“原因未查明”时，后文不要继续补充未经验证的解释。 这可能也解释了前文那个现象 → 该现象与截断的关系尚未验证 第十章“后来查明了机制”之后的内容已经得到确认，可以保留。 规则 7：不使用口语词 赢得很干脆 → 差值为 0.030 测得更准 / 测得更糙 → 估计精度更高 / 更低 可选的余地很小 → 候选范围小 基本上等于抓阄 → 接近随机选取 本来就分不出高下 → 真实差距低于可分辨范围 白跑 / 白占 → 无效运行 / 空占 不是白捡的 → 需要 79 GPU·小时 一个致命问题 → 主要问题 有事后找补的嫌疑 → 该切分方式在观察结果之后确定 原因不复杂 → 删除 既然预测这条路走不通，那就退一步 → 删除，直接进入实验设置 规则 8：不使用拟人表述 天生需要几百到上千次采样 → 该信号所需的采样量为几百到上千次 一旦超就制造出参差 → 截断发生时会增大奖励方差 24 步训练只挪动 0.05 → 24 步训练后正确率变化为 0.05 超得越彻底反而越稳定 → 截断率越高，奖励方差越低 回答内容本身的好坏参差 → 答案质量的差异查看被引原帖 ↗
Bilawal Sidhu@bilawalsidhu · 博主 · 6 天前
Who knew PMF for MediaPipe hand tracking would be performance art for social media :)
One of the unpredictable things about vibe coding is it breathes new life into tech that's been around for ages and gives it a new audience. Threejs is another example.
Top stories in AI today:
- AI labs head to the White House to discuss safety
- HeyGen founder replaced himself with an AI clone
- Build a project room where humans and agents work
- Finance execs find AI skills more valuable than MBA
Top stories in tech today:
- Silicon Valley’s push to sell experimental drugs
- SpaceX to nab Louisiana’s damaged coast
- Even more bad buzz for Flock’s AI cameras
- Apple steers buyers away from MacBook Air
- Quick hits on other tech news
九原客@9hills · 中文博主 · 6 天前
在大公司、政府机关干过的都知道，世界就是一个巨大的草台班子。
苹果法务部也不例外，太草台了。
人类组织成功的是少数，失败的是多数。能够带领着一帮草台班子还能把事做成，确实也不容易。
最近我对我的Agent们宽容了很多，因为我发现相比于之前对接过的某些人，Agent草台程度比他们低多了。
引用 OpenAI Newsroom @OpenAINewsroomApple is getting this wrong. openai.com/index/apple-is-ge…查看被引原帖 ↗
Bilawal Sidhu@bilawalsidhu · 博主 · 6 天前
Opus 5 is the like the dog that wreaks havoc in your office, eats all your homework, and then looks at you with an adorable face in the morning.
idoubi@idoubicc · 中文博主 · 6 天前
上线了一个扁平化的图片站模板：ShipAny Image Lite
核心功能就一个 ai image generator + 落地页，主打极速上线。有需要的可以看看👇
shipany.ai/zh/templates/imag…
Access offer:
app.runwayml.com/
九原客@9hills · 中文博主 · 6 天前
genoffice 还不错，但是不支持BYOK。官方其实代码里有BYOK的支持但是并没有做配置页面，所以我加了一个简单的配置页面。（源码下载后，让 Agent 帮你构建即可）
使用官方很简陋的实现，有空合并下 pi-ai 的逻辑，这样支持 codex 订阅等服务商。
github.com/ninehills/genoffi…
感谢，继续打磨中！
引用 NOW A LAKER @aegontara@yetone 大佬的alma真是一个越用越喜欢的产品，原本以为claude code和codex不断更新的情况下，alma会跟不上，没想到半年过去了，还是在用alma，最近出的折叠工具调用样式和光感执行过程太美观了，审美无敌！查看被引原帖 ↗
🎉 Congratulations to our Top 20 Love Story winners!
🎁 Each winner will receive 1,000 Credits.
Thank you for sharing your heartfelt stories with us! Rewards will be sent via DM within 5 days. Please check your inbox.
@musitsune
@Jenny_MommaLion
@falcothebard
@Malang6000
@OuXiaoxuan40782
@JeffreyJan5774
@tanhasami1122
@Simemeulation
@RealNovaSpark
@oceanscarr
@k_vti80903
@dianalfyu
@toyz_dez
@skystaa
@RamaLakshm53710
@Pippo265748257
@danu337368
@lexijfox
@ROCKStkl
@byaeryss
v0@v0 · 公司官方 · 6 天前
v0.app/docs/api/v2
Riley Goodside@goodside · 博主 · 6 天前
To illustrate this is a Blender render and not multimodal imagegen, an animation of the camera circling the board:
Read more:
therundown.ai/p/ai-giants-he…
Read more:
tech.therundown.ai/p/biohack…