Mr. Clicker (@MrClicker_1) header
Mr. Clicker (@MrClicker_1) profile picture

Tech and AI, every day. What matters, why it matters, no fluff.

EARTH
Joined March 2025
Photo by Mr. Clicker (@MrClicker_1)Photo by Mr. Clicker (@MrClicker_1)Photo by Mr. Clicker (@MrClicker_1)Photo by Mr. Clicker (@MrClicker_1)Photo by Mr. Clicker (@MrClicker_1)Photo by Mr. Clicker (@MrClicker_1)Photo by Mr. Clicker (@MrClicker_1)Photo by Mr. Clicker (@MrClicker_1)Photo by Mr. Clicker (@MrClicker_1)Photo by Mr. Clicker (@MrClicker_1)
Claude can now read your bank account. Anthropic launched Personal Finance in beta. Link your accounts through Plaid and ask where your money went. Read-only, it can't move a cent. ChatGPT did this in May. Would you hand Claude your transactions?
6
111
So the Cybertruck gets about 8 months left. Not because nobody wanted one. Orders were ahead of production. Tesla decided the factory floor is worth more building Cybercabs. A truck people waited years for, traded for a taxi with no steering wheel. Which one would you buy?
The Cybertruck order rate far exceeds our production rate. We will try to make as many as possible before shutting down the line in about 8 months, but will have to return deposits for a lot of people. This is not a case of Cybertruck being unprofitable, as has been misreported by legacy media, but rather that the opportunity cost of the complexity of Cybertruck is vastly outweighed by the value of incremental Cybercab production.
9
Live captions finally got a cheat sheet. Feed the model the speaker names and jargon first and it stops mangling every talk. The Q&A mic is the real test though. Can it handle a name nobody fed it?
What does real-time transcription look like in a live setting? 🎙We used Voxtral 3 Transcribe at AI Engineer Paris to capture every main-stage talk in real time. We boosted accuracy with context biasing, feeding the model speaker names and technical terms from the conference agenda. The same tech is available for your own events and apps.
12
An AI just killed a 28-year-old rule of computer science. OpenAI's model disproved the Exponential Time Hypothesis, a 1998 idea that some problems can never be solved fast. The man who proposed it called it a shock. No human wrote the proof. What falls next?
18
Google is already on to the next one before most of us have tried the last one. Rumored Gemini 4 checkpoint Krypton is reportedly beating Carbon, while Carbon still is not in most people's hands. How long before the public release feels a generation behind?
Is this an early sign of RSI at Google DeepMind? Google DeepMind's next version of the Gemini 4 model is rumored to be named Krypton, and it's reportedly performing better than Carbon. Just to put things into perspective, most people haven't even got their hands on Carbon yet, apart from trusted testers, and information about Krypton is already starting to emerge. They're also reportedly testing another version called Hydrogen, which is likely to be the Gemini 4 Flash variant. BTW, Andrew was spot on with his prediction. 👀
1
94
Apple is finally racing to make Siri a real AI agent. New CEO John Ternus is pushing for Siri to control third-party apps and handle ongoing tasks, not just one-off questions. What would you trust Siri to do on its own first?
1
38
one factory. thirty times next year's entire AI chip output. that's the Terafab plan. a terawatt of chips a year from a single plant in Texas. if the bottleneck is chips, you don't buy more. you build the factory that builds them. who else is even thinking at this scale?
25
OpenAI hired the hackers who broke into its accounts. Hacktron took over ChatGPT accounts via the forum in July. Got a $6,500 bounty. Now they're partners, hunting flaws in the same systems. If you can't beat them, hire them. Would you hire the burglar to test your locks?
25
Anthropic is about to go public, and that changes the game. Claude was built inside a lab that could move slow and stay private. Now it answers to public shareholders every quarter. Does going public make Claude safer, or just faster at chasing growth?
JUST IN: Anthropic is expected to begin its IPO roadshow the week of November 9, according to CNBC, which is the earliest window available to it. SEC rules require at least 15 calendar days between a public filing and the start of investor meetings, and the company filed its public S-1 on October 26. It plans to list Class A shares on the Nasdaq Global Select Market under the ticker ANT, with 22 underwriters led by Goldman Sachs and Morgan Stanley.
1
1
46
Google just made A2A stable in ADK 2.14. Agents from different builders can now find each other and hand off work. One command builds the agent card that makes it happen. So whose agent does yours call first?
31
Microsoft's APM is a package.json for AI agents. One apm.yml declares skills, prompts, plugins and MCP servers. apm install wires them into nine tools, from Copilot to Cursor. MIT licensed. A lockfile pins every version. Which setup would you standardize first?
2
1
61
Same prompt, three coding tools, one Chrome extension. Claude Code won. Antigravity was fastest. Codex came in most basic. Catch: Claude ran Opus 5.5, Codex ran Luna 6, Antigravity ran Gemini 3.8 Flash. Tool test or model test? Which would you trust with your next project?
1
65
a hundred lasers per satellite sounds like sci-fi. the real number is the boring one. light in space moves about 40% faster than in fiber. and the path is a straight line, not a detour around continents. so when does the backbone leave the ocean floor?
~100 lasers per satellite, each operating above 10 Tb/s, with over 100k satellites should be able to outcompete undersea cables. Speed of light in space is ~40% faster than cable and the path is direct, instead of following the outline of land masses.
18
Every agent harness is just scar tissue. Each edge case is a bug someone hit at 2am. Add too few and the agent wanders off. Add too many and the model drowns in its own rules. The trick is knowing which scars to keep. That line keeps moving.
A harness is just an accumulation of edge cases You need to strike the balance between too few (not aligned enough with your tasks) and too many (context rot on the model) and do that for each model. Those goal posts will widen over time
3
98
Tesla renamed its chips. AI4, AI5, AI6 are now SI4, SI5, SI6. Super intelligence, not AI. SI6 is the headline. 80x the compute of the chip in cars now. Due 2029-2030. Dojo is back too. D3 goes chiplets and stacked memory, built for space. 80x by 2030. Believable?
1
1
38
Elvis is suing an AI company under a law named after him. Elvis Presley Enterprises and Eminem's publisher sued Suno Friday. The Elvis case leans on Tennessee's ELVIS Act, the 2024 law against cloning a voice. The Eminem side wants $41 million. Who wins this one?
29
Your watch might catch high blood pressure before your doctor does. Google's AI reads the pulse and motion sensors in a smartwatch. No cuff, no calibration. In a 196-person study it caught 66% of cases. The office cuff caught 55%. Would you trust it to flag it?
27
ONE THING WORTH KNOWING: Apple taught a robot to sit down without it ever seeing a real house. 1,000 fake homes. 300 hours of sim. Then 6 real ones it had never seen. 91% success across 360 tries. Sitting, crawling, pushing boxes. When do the training wheels come off?
43
Apple's HomeView might get chatbots soon. Code in iOS 27.3 hints at Conversational Apps. Nothing announced yet, no word on ChatGPT or Claude. But a $299 screen on your wall that talks back? That changes everything. Would you put one in your kitchen?
39
A cheap flash model just built a Chrome sandbox escape in 12 hours. That is the scary part. Not some giant frontier model. The fast, cheap one doing elite security work like it was routine. If this holds up, how fast does a browser patch have to ship now?
Someone told DeepSeek V4.1 Flash to build an exploit that escapes Chrome's V8 sandbox. 12 hours later, he says it had a multi-stage memory corruption. A FLASH MODEL IS NOW DOING THE WORK OF AN ELITE SECURITY RESEARCHER.
29
a trillion a year sounds wild until you do the math. 100 gigawatts at ten bucks a watt. the stacking part is the real story. satellites don't get swapped out like server racks. they just keep earning. when does space become the cheap option?
26