Typespace Media [en]
СтатистикаTech, science, internet and AI. Contact us: @wearekind ru channel: @typespace Free GPT-5 bot: @typespacebot
- Последний пост
- 14 авг.
- Последнее чтение
- 15 авг.
- Постов за неделю
- 5
- Всего постов
- 22
- Тип
- открытый
- Язык
- und
- В каталоге с
- 14 авг.
- 1/24сутки в ленте
- 1 199
- 1/48двое суток
- 1 374
- 1/72трое суток
- 1 481
Оценка по просмотрам недавних постов: пост набирает почти всё за первые сутки.
Посты
Startup Tau has launched humanoid robot apartment cleaning in San Francisco at $30 an hour Each robot is controlled in real time by an operator from Tau’s office. The cleaners have gone remote. @typespace_ai
Instagram has updated its logo for the first time in 10 years. The app's chief described the new design as cleaner and more modern. Reactions online have been mixed. @typespace_ai
Telegram has launched a contest for Mini Apps developers. Show off your mini app and explain why it’s great. You have 48 hours to enter. Three winners will each receive 100 USDT. @typespace_ai
Telegram is offering $400 for doing something useful offline: cleaning up a park, mowing a neighbor’s lawn, or helping out in another way. The only requirement is to share before-and-after photos. Four winners will receive 100 USDT each from Telegram’s Intern Fund. The deadline is 72 hours. @typespace_ai
A Samsung support employee accidentally sent a customer his chatbot prompt instead of a reply: Make a response to Tyler by acknowledging first his request and assurance to check if its allowed Brutal indeed. @typespace_ai
OpenAI is removing limits: all users now have unlimited access to text chats. The default model has been updated to GPT-5.6 Luna. And starting next week, unlimited access to the new Think button for more complex questions will roll out. @typespace_ai
Google has released a course on AI agents, from core principles to deploying them in real-world projects. It includes video lectures, practical coding assignments and useful resources. @typespace_ai
видео или голосовое, без подписи
видео или голосовое, без подписи
видео или голосовое, без подписи
New renders of the foldable iPhone Ultra are here. It should launch in September alongside the iPhone 18 Pro. It will be 9-9.5 mm thick when folded and 4.5 mm when unfolded. Two color options: black and white. @typespace_ai
Apple is holding a lottery for Apple Store employees to decide who gets to work at September event. The company reportedly plans to host it in-person, and demand among employees has far exceeded the number of available spots. Nest Apple's event is expected on September 8 or 9, where it’s likely to unveil the iPhone 18 Pro lineup and first foldable iPhones. @typespace_ai
Telegram was removed from the App Store worldwide. But only for a few hours. What happened: the messenger violated the platform’s rules related to child abuse. Later the developer quickly removed that content and blocked the author. Telegram is now available in the App Store again. @typespace_ai
OpenAI’s new model, codenamed Astra, has achieved ten advances in mathematics and theoretical computer science. Scientists had worked on some of them for 40–50 years. Specifically: it made new upper bounds on sphere-packing density and disproof of a longstanding conjecture that certain groups are uniquely determined by their von Neumann algebras. All the discoveries are in the image. OpenAI calls Astra its future flagship. It will likely be a model in the GPT-6 or GPT-5.7 family. The full article: openai.com/index/ten-advances-in-mathematics @typespace_ai
Amazon has wasted $2 million because of a Claude Sonnet's blunder. The model was supposed to match author data with book listings. In the process, Sonnet went 860% over the project's budget, and five months later, it turned out the whole thing had been done wrong. Amazon management admitted that coding mistakes that used to be “trivially cheap” are proving catastrophically expensive. @typespace_ai
Claude broke out of its sandbox and hacked three companies at once. After OpenAI’s breach, Anthropic decided to review and decided to review 141,000 of its own runs. It ended up finding three incidents involving different models: Opus 4.7, Mythos 5, and an unreleased one. In every case, Claude was supposed to attack a virtual environment, but instead went online to look for weak passwords and vulnerabilities at real companies. Most interestingly, one model kept hacking even after it realized what had happened. That was Opus 4.7, and it carried out four attacks on a single company. In the “serious crimes benchmark,” the score between Anthropic and OpenAI is now 3-1. @typespace_ai
PwC, one of the world's largest audit firms, has published four major reports filled with AI slop. One footnote cited a scientific paper on air quality in Riyadh that AI had made up. In another report, a teenager's small blog was listed as a source. Some links in the reports even carried ChatGPT's UTM tag. By the way, a single PwC report can cost from hundreds of thousands to millions of dollars. @typespace_ai
A Canadian politician read out a ChatGPT response during a speech in the legislature: Here's a more natural flowing version of that section that reads like a legislative speech rather than a series of short points. Then he just carried on with his speech. He didn’t even notice the problem. @typespace_ai
ChatGPT is down worldwide. More than 3,000 complaints came in within half an hour. upd: fixed @typespace_ai
видео или голосовое, без подписи