Chai with AI โ˜• (@ChaiWithAi) header
Chai with AI โ˜• (@ChaiWithAi) profile picture

Brewing the biggest AI & tech drops โ˜•๐Ÿš€ Building @bot https://teapawt.org/t.co/d2fLT8JoDO Remote jobs: https://teapawt.org/t.co/5t8ce0wVCP

Joined September 2015
Photo by Chai with AI โ˜• (@ChaiWithAi)Photo by Chai with AI โ˜• (@ChaiWithAi)Photo by Chai with AI โ˜• (@ChaiWithAi)Photo by Chai with AI โ˜• (@ChaiWithAi)Photo by Chai with AI โ˜• (@ChaiWithAi)Photo by Chai with AI โ˜• (@ChaiWithAi)Photo by Chai with AI โ˜• (@ChaiWithAi)Photo by Chai with AI โ˜• (@ChaiWithAi)Photo by Chai with AI โ˜• (@ChaiWithAi)Photo by Chai with AI โ˜• (@ChaiWithAi)
966 remote jobs. Some pay up to $450 an hour. I built a free site that puts open roles from Mercor, micro1, Handshake AI and Babel Audio in one place. โ€ข No sign-up to browse โ€ข Search by role, see pay upfront โ€ข About 400 roles pay $100+/hr โ€ข Doctors, lawyers, engineers, writers, voice work I was tired of checking multiple sites every day, so I made one. jobs.letsremote.here.now What role should I add next? #RemoteJobs #AI
9
9
31
821
A 99-out-of-100 game, a deadline for lazy foldable apps, and a rocket failure with a first answer. ๐ŸŽฎ GTA 6 reviews are in, and they are near perfect After 13 years of waiting, Grand Theft Auto VI holds a 99 on Metacritic, tying The Legend of Zelda: Ocarina of Time as the highest-rated game ever, with over a hundred critics and not one mixed or negative review. It unlocks at midnight local time around the world, and preorders are estimated at $1.54 billion. Why it matters: the most hyped game in history actually lived up to it, and tonight it finally reaches players. ๐Ÿ“ฑ Google Play sets a foldable deadline for apps From April 2027, Android apps must fill the whole foldable screen with no black bars and keep your place when you fold or unfold, or they will not be allowed to publish updates on Google Play. Listings will also have to show foldable screenshots, and games get until August 2027 to meet the same bar. Why it matters: foldables stop being second-class citizens, and sloppy apps will simply stop updating. ๐Ÿš€ ULA points to fuel supply in the Vulcan failure Preliminary data from the November 15 failure that destroyed 40 Amazon Leo satellites points to inadequate fuel supply reaching the BE-4 engines, not the engines themselves, according to ULA. The investigation continues under the FAA, and Amazon's next Leo launch is an Ariane 6 flight targeted for November 23. Why it matters: clearing the engines keeps Blue Origin out of the blame frame, but Amazon's satellite race against Starlink just lost precious time. Reviews, rules and rockets: tonight's tech news spans the biggest game ever, the future of foldables, and a setback in orbit. Which of the three are you watching closest? #TechNews #GTA6 #Android #Space
1
3
208
Chai with AI โ˜• retweeted
You: Want a PS5 to play GTA 6. We: Are giving 6 away. Just repost this, reply with why you want it, we'll pick the best replies to win. Paid out in upgraded money (USDC). US only.
4,135
4,009
523
5,247
346,069
Chai with AI โ˜• retweeted
Giving away a free copy of GTA 6 To enter: - Repost this post - Reply with Xbox or PS5 Winner announced in my Discord before midnight ET
8,836
9,371
356
17,967
915,176
Google's top AI plan is not a subscription. It is an entire toolbox. ๐Ÿงฐ Here is what Google AI Ultra actually gives you: ๐Ÿค– Gemini, maxed out 5x higher usage limits than the AI Pro plan, access to Gemini 4 Carbon, Deep Research, and the highest limits in Google AI Studio. Gemini also built into Gmail, Docs, Sheets, Slides and Vids, plus the highest access inside Google Search. ๐ŸŽฌ Create almost anything Google Flow for cinematic scenes and stories, Flow Music, Nano Banana Pro, and Project Genie, where you build and explore interactive worlds in real time with the Genie 3 world model. ๐Ÿ’ป Built for developers Google Antigravity with higher limits on Gemini 4 Carbon and Opus 5.5, Gemini in Android Studio, and Google Developer Program premium with higher Cloud credits. ๐Ÿ–ฅ๏ธ Run models on real GPUs Google Colab is included: 1,000 compute units a month (2,000 on the 20x plan), worth about $100 if you bought them separately. That is roughly 840 hours on a T4 GPU, 580 hours on an L4, or 185 hours on an A100, plus Premium GPU access, longer runtimes, and background execution, so your notebook keeps running even after you close the tab. Enough GPU time to run open models like the ones I post here every day. ๐Ÿ“š Research and learn Expanded NotebookLM with Audio and Video Overviews, Quizzes and more. ๐Ÿ  The everyday stuff 20 TB of storage across Photos, Drive and Gmail, shareable with up to 5 people. YouTube Premium Individual included. Google Home Premium Advanced with 24/7 video history, and Google Health Premium if you pair a Pixel Watch or Fitbit. Some limits: this is the premium tier above AI Pro, so expect a premium price. A few perks also need extra hardware, like a Pixel Watch for Health Premium or Nest cameras for Home Premium. Colab's per-hour rates are not officially published and can change, and no specific GPU is guaranteed. One plan for your chat, your code, your videos, your files, your home, and now your GPU. Tell me what should i build with this? #GoogleAI #Gemini #AI
401
OpenAI just gave ChatGPT two new brains, and the cheaper one nearly beat the flagship. ๐Ÿš€ ๐Ÿง  GPT-6.2 Astra: the new flagship Astra is OpenAI's model for your hardest projects. Show it a screenshot of one slide and it rebuilds the rest of your deck to match, down to the exact font size. It writes cleaner code, tops OpenAI's Agents' Last Exam for long professional tasks, and its API price just dropped 20%, to $8 in / $40 out per million tokens. โšก GPT-6.2 Sol: the smart bargain Sol is the everyday model, and on Cognition's FrontierCode 2.0 test, where real code changes are judged on whether maintainers would merge them, Sol finished within half a point of Astra at about one-fifth of the cost. It even beat the old GPT-6 Astra on OpenAI's LifeVal test of everyday tasks, at about one-seventh of the cost. Same price as GPT-6.1 Sol: $2 / $10 per million tokens. ๐ŸŽ๏ธ Ultrafast mode Both models can now generate up to 4.5x faster in Ultrafast mode, though it burns more of your allowance. The $200 Pro plan gets Ultrafast on Sol for the first time, and the $500 Pro plan gets it on both models. Both are live now in ChatGPT Work and Codex on Plus, Pro, Enterprise and Edu, and through the API. OpenAI says both also finish demanding tasks with fewer output tokens, so your usage limit stretches further. All the benchmark numbers above are OpenAI's own. A flagship that got cheaper, and a budget model that nearly caught it. Which one would you run your daily work on? #ChatGPT #OpenAI #GPT6 #TechNews
99
The last helicopter has landed. ๐Ÿš๐Ÿ’› MS Dhoni has retired from the IPL. No press conference. No farewell tour. On a routine retention call, as CSK read out the list of players being released, Dhoni quietly said: "Maybe you should add my name in it." That is how the greatest IPL career ever ended. In one quiet sentence. The numbers behind the silence: ๐Ÿ† 5 IPL titles as captain: 2010, 2011, 2018, 2021, 2023 ๐Ÿ 278 matches, 5,439 runs, the only captain with 100+ IPL wins ๐Ÿ“… In the league since day one in 2008, the last active player from that first season He missed the entire 2026 season injured. At 45, the body decided before the heart could. But he is not leaving Chennai. CSK has named him Team Mentor for IPL 2027. The dugout keeps its brain. The pitch loses its finisher. In 2020 he said goodbye to India with one Instagram post. Tonight he did it again, with the same song. Some players retire. Dhoni just logs off. Thank you, Thala. ๐Ÿ’› What is your one favourite Dhoni IPL moment? #Dhoni #CSK #IPL #ThankYouDhoni
92
Tuesday's tech: the world's biggest websites served malware to their own visitors, Gemini learned to speak like a local, and Xiaomi's thinnest band went global. ๐Ÿ›ก๏ธ The script that hijacked 178 famous websites Attackers broke into Optimizely's A/B testing tool and slipped malicious code into the scripts loaded by CNN, the BBC, Zoom, HSBC and 173 other big sites. Visitors got a crypto wallet swapper, a card skimmer, and on older iPhones an exploit kit. Optimizely says it fixed the flaw and removed the code, and researchers counted about $7,040 in stolen crypto. ๐ŸŽ™๏ธ Gemini now speaks in 66 local voices Google is rolling out 66 regional voices for the Gemini app on Android and iOS, including Indian English and Hindi plus UK accents from Geordie to Scouse. Gemini Notebook can also now turn your notes into interactive websites, like a study timeline or a sales dashboard. โŒš Xiaomi's thinnest band goes global for โ‚ฌ59.99 The Smart Band 11 is now on sale in Europe and the UK ahead of tonight's Paris event, with a 1.72-inch AMOLED screen at up to 2,000 nits and up to 21 days of battery in light use. It tracks heart rate and blood oxygen around the clock and covers 150-plus sports modes. A supply-chain attack, an AI that finally sounds like home, and a budget band that lasts three weeks - Tuesday's tech ran from alarming to oddly comforting. Which story grabbed you most: the hijacked websites, Gemini's new voices, or the budget band? #TechNews #Optimizely #Gemini #Xiaomi
116
Tuesday's tech: a phone with a screen on its back, a $38 billion courtroom wait, and a CRM that fills itself in. ๐Ÿ“ฑ Xiaomi's flagships go global tonight Xiaomi takes the 18 Pro and 18 Pro Max worldwide at a Paris event at 6:30 PM IST, two months after their China debut. The draw: a second display on the back, dual 200MP Leica cameras and Snapdragon's new 8 Elite Gen 6 chips. The Watch S5, Smart Band 11 and Clip earbuds join them on stage, but global prices are still under wraps. ๐Ÿ’ณ Visa and Mastercard wait on a $38B verdict A US judge heard arguments on Monday over the card giants' swipe-fee settlement with merchants, but didn't rule, warning that rejecting it could mean a long, drawn-out trial. The deal would cut fees by 0.1 points for five years, cap standard card fees at 1.25% for eight years, and let shops refuse some premium cards for the first time. The $38 billion is the estimated saving for merchants by 2031, not a payout, after US merchants paid $118.8 billion in swipe fees last year. ๐Ÿค– Freshworks veterans raise $53M to rebuild the CRM Dreamteam, founded by three former Freshworks executives, has raised $53 million from Accel, Together Fund and Premji Invest for a CRM that updates itself from emails, calls and meetings. Its Company Brain and four AI teammates do the admin salespeople hate, and it already has 100-plus customers after three months of selling. A phone that shows its back, a courtroom that could rewrite every card swipe, and a CRM that does its own paperwork โ€” Tuesday is about tech that stops asking you to do the boring parts. Which one would change your day most: the phone, the card fees, or the self-filling CRM? #TechNews #Xiaomi18Pro #Dreamteam
203
Free today: no new model dropped overnight, but the open-source toolbox grew - a self-hosted AI on-call engineer, a free way to run your own inference cloud, and a spell-checker for your AI agent's instructions. Free tools โ€ข stromboli (OpenStatus): a self-hosted AI SRE agent that jumps on alerts, suggests a fix first, then hunts the root cause in your logs and metrics and reports to Slack โ€” MIT licensed, works with Claude, OpenAI, Gemini, DeepSeek or local models โ€ข Kinference (Buoyant): an open-source control plane for running AI inference on Kubernetes, built on vLLM, with routing, guardrails and per-team cost tracking โ€” Apache 2.0 โ€ข AsItIs Agent Tools: a free MIT-licensed linter that checks the CLAUDE.md, AGENTS.md and SKILL.md files your coding agent reads, catching broken imports, hidden instructions and token-budget blowouts before they reach production Some limits: stromboli is brand new and needs your own LLM key, so the agent is free but the model calls are not; Kinference's speed claims (95% faster first token, 45% lower latency) are Buoyant's own benchmarks; AsItIs is a niche developer tool, not a general AI app. Which of these would you actually run: the on-call agent, your own inference cloud, or the instruction linter? #FreeAI #OpenSource
4
1
1
146
The two biggest AI money moments of the week land back to back, and both hinge on trust. ๐Ÿ’ฐ Nvidia reports today, its first test without Jensen Huang Nvidia reports Q3 earnings today after the US close, the first results since Huang resigned on 12 November. CFO Colette Kress is interim CEO. Wall Street expects about $2.47 a share on about $109 billion in revenue, just above Nvidia's own $108 billion guidance. The numbers matter, but the call matters more: what Kress says about the next quarter, the CEO search, and the export questions Treasury Secretary Scott Bessent raised, with board answers due within 10 days. Update: CNBC reported Monday that the board is not considering Huang's return and is focused on its investigation, despite Trump saying he would not mind and Musk backing a comeback. Shares fell 4.1% on Monday. ๐Ÿš€ Anthropic prices tomorrow, then starts trading as ANT The Claude maker is expected to price its IPO on Wednesday and start trading on Nasdaq as ANT on Thursday. The range is $70 to $78 a share for 825 million shares, which could raise up to about $64 billion and value the company near $1.4 trillion. Its filing shows $33 billion in revenue in the first nine months of this year, up from $2.5 billion a year earlier. Orders were reported at about six times the shares on offer, and Altimeter Capital disclosed a $5 billion order on Monday. The order book closes today. One chip giant proving it can run without its founder, one AI lab proving the market still believes. Which one are you watching first? #TechNews #Nvidia #Anthropic #IPO
56
The most powerful chipmaker in the world might get its old boss back, the biggest game launch of the decade has a box problem, and a chat message just hacked a game. Monday night is loud. ๐ŸŸข Jensen Huang is weighing a comeback at Nvidia Four days after resigning as Nvidia's CEO, Jensen Huang has discussed returning with confidants, the Financial Times reports, while allies including Elon Musk build a "war room" to lobby the board. Treasury Secretary Scott Bessent wants a dramatic shake-up instead, and the board's investigation is still running. Nothing is decided, but the chipmaker that powers the AI boom is now a boardroom tug-of-war. ๐ŸŽฎ Some GTA 6 boxes arrive with no working code inside Days before the November 19 launch, buyers report physical GTA 6 copies arriving with a blank strip where the download code should be, or a code already redeemed. With no disc in the box, those buyers cannot install the game at all. Rockstar has not confirmed how widespread it is, and some retailers are refusing replacements while others help. ๐Ÿ’ฌ A chat message hacked a free Steam game An attacker sent a crafted direct message in CivIdle that slipped malicious scripts through the game's chat notifications onto Windows PCs. The developer fixed it server-side within hours, disabled DMs indefinitely, and told players to check for two files, q.exe and p.bat, and update to Build 1018. Linux and Mac players were not affected. A comeback, a code that isn't there, and a chat that bites back. The week before GTA 6 might be the loudest week in tech this year. Which of these three would you least want to deal with tonight: a boardroom fight, a blank game box, or a hacked chat? #TechNews #Nvidia #GTA6 #Gaming
1
1
227
China's biggest AI name just shipped its new flagship behind closed doors, Washington is picking its AI champions, and a Linux bug from 2015 can still crash your server. ๐Ÿค– Alibaba's Qwen4-Max is here, but you can't download it Alibaba launched Qwen4-Max, its new flagship AI model, with a 1-million-token context window and an API price of $1.50 per million input tokens. Alibaba claims it beats Anthropic's Claude Fable 5.1 on coding benchmarks, though those are the company's own numbers, not independently verified. The catch: the weights are closed, a break from the Qwen models developers know. Alibaba says smaller Qwen4 models will be open, but gave no date. ๐Ÿ† Washington is about to name its AI "national champions" The Trump administration plans to designate more than two dozen US AI companies, including Nvidia, Amazon, Microsoft, OpenAI and Anthropic, as "national champions" to help them expand overseas and compete with Chinese rivals, the Wall Street Journal reports. Commerce Secretary Howard Lutnick is expected to announce the list within about a week. The report is unconfirmed, and Reuters says it could not verify it. ๐Ÿง A Linux bug from 2015 can crash a server without logging in Researchers disclosed NoisyTCP, an 11-year-old flaw in the Linux kernel's TCP code tracked as CVE-2026-99288. An unauthenticated attacker can crash a server remotely, though no remote code execution has been shown and no attacks in the wild are reported. Fixes are in kernel versions 6.12.117, 6.18.63 and 7.3.4, so patching is the move. ๐ŸŽฎ The next Xbox is a console that plays your PC games, not a PC Microsoft's 25th anniversary magazine gave the clearest picture yet of Project Helix, its next Xbox. Xbox chief Jason Ronald says it is "a console that is also capable of playing games from your PC," keeping four generations of Xbox games playable, with PC tinkering optional. No price or launch date yet, and Microsoft admits the component crisis is shaping how the console gets built. ๐Ÿ›ฐ๏ธ Samsung is set to build Starlink chips in Texas Samsung plans to make the STM32V8 microcontroller, the chip that runs laser links between Starlink satellites, at its Austin fab, as part of a $1.9 billion expansion. Its newer Taylor plant stays focused on advanced AI chips, including Tesla's. Production is targeted for 2030, and Samsung won't officially confirm individual customer contracts. A closed flagship, a champions list, a patched bug, a console that eats your PC library, and satellite chips made in Texas. Monday is moving fast. Which of these five will matter most a year from now? #TechNews #Qwen #Nvidia #Linux #Xbox #Samsung
2
1
2
194
Apple wants 200,000 iPhone owners to ask an AI about their health, and Kratos just showed up on a 13-year-old console. ๐ŸŽ Apple opens a 200,000-person study on AI health answers Apple and OpenEvidence have opened enrolment for the Apple Health AI Usage Study, a research app where US iPhone owners ask everyday health questions and get plain-language answers cited to journals like the NEJM, JAMA and Nature. It is open to 200,000 people aged 21 and over with an iPhone 15 Pro or later on iOS 26. Apple says the data is anonymised, never trains AI models, and OpenEvidence deletes queries after 24 hours. ๐ŸŽฎ God of War: Sons of Sparta reaches PS4, with two brutal new modes The 2D God of War prequel that launched on PS5 in February arrives on PS4 at midnight local time, and a free patch adds two new modes to both consoles. Hermes Mode is built for speedrunners, with a split timer and a persistent scoreboard. Erebus Mode is the opposite: one hit kills you, and one hit kills your enemy. One giant experiment in trusting AI with your health, and one old console getting a new lease on life. Monday is not easing in gently. Would you let an AI answer your health questions if researchers were studying how you used it? #TechNews #Apple #PlayStation #GodOfWar
52
Free today: a 188-billion-parameter brain you can download and run yourself. Free models โ€ข Agnes 2.5 Pro Theta (Agnes AI): a multimodal open-weight model that chats, reasons, codes and reads images and video, with a 262K-token context and an Apache 2.0 licence Some limits: the weights are about 350 GB, so this is datacentre hardware, not a laptop, and its benchmark scores are the company's own, not independently verified. It also needs custom code to run, so inspect that code before trusting it. Would you run a model this big yourself if you had the hardware? #FreeAI #OpenSource
2
1
104
One chip scandal deepens, two governments rethink Big Tech, and AMD's next chip is already leaking. ๐Ÿ” The chip scandal's next chapter Megaspeed, the Singapore cloud firm at the centre of the Nvidia chip-diversion probe, has hired outside lawyers to investigate media reports about its AI cloud business and supply chain. It says the reports lack verifiable evidence and that former executive Alice Huang left in August 2025. The move follows Jensen Huang's resignation as Nvidia chief and Congress lining up December hearings. Why it matters: the probe now spans a company, a resignation and two countries' investigations, and the chip at the heart of the AI boom is also at the heart of a national-security row. ๐Ÿ‡ต๐Ÿ‡ฑ๐Ÿ‡ง๐Ÿ‡ท The outage that changed policy Poland plans to expand open-source software across its public administration after the Microsoft 365 outage that hit 74 countries, starting with pilot projects in local government. In Brazil, a lawmaker is using the same outage to push a digital sovereignty amendment. Why it matters: a 36-hour outage is now reshaping how governments buy software, not just how IT teams patch it. ๐Ÿ’ป AMD's mystery chip, leaked Reports say AMD is working on a processor called "Siren Halo" that would pair low-power LPDDR memory with separate GDDR7 graphics memory, an unusual split for an integrated chip. AMD has not confirmed the name, the specs or the 2028 timeframe the reports float. Why it matters: if real, it would narrow the gap between built-in graphics and a separate graphics card in small PCs. One weekend, three reminders: supply chains are political, outages are policy, and the next chip is already a rumour. Which story are you watching closest? #TechNews #Nvidia #AMD #Microsoft
68
Why doesn't Tesla have this feature yet? @elonmusk
China is so far ahead ๐Ÿ˜ญ
1
99
Free today: run Claude Code on Kimi and Codex on DeepSeek, all from your menu bar. Free tools โ€ข Magpie (yetone): free, MIT-licensed menu-bar app, 36K GitHub stars, that lets any coding agent run on any model, with a local gateway, automatic failover when a quota runs out, and on-device masking that keeps your secrets off vendors' servers Some limits: Magpie itself is free, but you still pay for the model calls or the subscriptions behind it. Privacy masking is off by default, and it reads your agents' sign-in files to find subscriptions, though you can switch that off. Which agent would you point at a cheaper model first? #FreeAI #OpenSource
2
1
123