Cory Chesebro

Seems like a pretty big blunder from Anthropic to do something like this without a unified replacement available. I understand not wanting to support a competitors product, but this is a user hostile move.

@curiousity_path @yacineMTB For sure it's useful, but when you break down into small enough tasks they become simple. If you can't break it into small tasks, it's not simple. Like a lot of research tasks or unexplored optimizations.

@itestinprd @yacineMTB The best use case that I found for claude so far, is that if you have a large workload of small tasks of a similar nature and you’re able to come up with a strict architecture and validation flow, you can save a lot of time, and I mean months.

If you are doing anything serious you very quickly see the limits of these new AI models

@lydiahallie Will there be usage rollbacks?

We're aware people are hitting usage limits in Claude Code way faster than expected. Actively investigating, will share more when we have an update!

@trq212 @altryne Any update on the leading theory? I am 90% of the way through my weekly limit when normally I would only be around 40% by now.

@altryne we are actively looking into this in particular but not sure it's real yet, prompt cache bugs can be quite subtle

PSA: If you've been running out of Claude session quotas on Max tier, you're not alone. Read this. Some insane Redditor reverse engineered the Claude binaries with MITM to find 2 bugs that could have caused cache-invalidation. Tokens that aren't cached are 10x-20x more expensive and are killing your quota. If you're using your API keys with Claude this is even worse. This is also likely why this isn't uniform, while over 500 folks replied to me and said "me too", many (including me) didn't see this issue. There are 2 issues that are compounded here (per Redditor, I haven't independently confirmed this) : 1s bug he found is a string replacement bug in bun that invalidates cache. Apparently this has to do with the custom @bunjavascript binary that ships with standalone Claude CLI. The workaround there is to use Claude with `npx @anthropic-ai/claude-code` 2nd bug is worse, he claims that --resume always breaks cache. And there doesn't seem to be a workaround there, except pinning to a very old version (that will miss on tons of features) This bug is also documented on Github and confirmed by other folks. I won't entertain the conspiracy theories there that Anthropic "chooses" to ignore these bugs because it gets them more $$$, they are actively benefiting from everyone hitting as much cached tokens as possible, so this is absolutely a great find and it does align with my thoughts earlier. The very sudden spike in reporting for this, the non-uniform nature (some folks are completely fine, some folks are hitting quotas after saying "hey") definitely points to a bug. cc @trq212 @bcherny @_catwu for visibility in case this helps all of us.

@status_effects Salty Bet for AI models?

llms can FIGHT now. here's opus as wizard vs gpt-5.4 as robot. calling this budok-ai. it works by modding the brilliant game yomi hustle. 8-model seeded tournament incoming. details and code below:

@RedPandaKoala Sooo... Did anything come out of this?

🚨 The hacker group Handala Hack claims to have hacked Lockheed Martin and in their telegram claim they will be releasing information tomorrow “Those who move in the shadows will soon see the light. The secrets hidden behind the curtains of smoke and mirrors are now ready to be revealed. Those who seek the truth will soon face a reality they have only seen in their nightmares.”

Look-ahead bias sucks

@WuBlockchain What's the point of this though? What are you going to use it for?

Strategy (MSTR) would need to maintain an average purchase rate of about 6,158 BTC per week over the remaining ~42 weeks of 2026 to reach 1 million BTC holdings by year-end. At an average price of $85,000 per BTC, it would still need to acquire 261,269 BTC, costing roughly $22.2 billion. As of last Monday, the company held 738,731 BTC. In 2026 year-to-date, Strategy has purchased 64,948 BTC, including 17,994 BTC added last week.

@joshuaday Yeah I have it hooked up to telegram, but I want to be able to modify the configs and rescue it if it gets stuck (it's happened a few times)

there's an android companion app that pairs with your gateway running on a server. the phone acts as a node (camera, notifications, location) but the gateway still runs on your mac/linux/VPS. you can also just use telegram or discord as the interface from any phone and skip the app entirely

Anyone got OpenClaw running on their phone yet?

POV: Senior Agentic Engineer

The missile is reading the Earth like a blind person reading braille. Running its fingers across the surface and matching what it feels against what it memorized. Before GPS, engineers had a problem. You’ve got a cruise missile flying 500 mph at treetop level toward a target 1,500 miles away. No pilot. No satellite link. How does it know where it is? Their solution: give the missile a topographic map and let it read the ground. TERCOM (terrain contour matching) works like this. Before launch, you load the missile with a series of altitude profiles of the terrain along its flight path. Think of it as a cross-section of every hill, valley, and plateau between point A and point B, sliced into strips. During flight, a radar altimeter on the belly of the missile pings the ground constantly. It measures the exact altitude of the terrain below and builds a real-time profile. Then the onboard computer slides that real-time profile across the stored map, looking for the best match. When it finds the match, it knows exactly where it is, and corrects course. The wild part: this worked over enormous distances in the 1970s. The Tomahawk cruise missile could fly 1,000+ miles through a pre-programmed corridor of terrain strips, checking its position every few minutes, and arrive within meters of a target. No GPS. No external signal. Just ground texture. Then it got crazier. Engineers added DSMAC, digital scene-matching area correlation. Same concept, but with a camera instead of a radar altimeter. In the terminal phase, the missile takes a photograph of the ground below, digitizes it, and compares it to a stored satellite image of the target area. Pixel by pixel matching. In the 1980s. The engineering constraint that made all of this necessary is the interesting part. Inertial navigation systems drift over time. Gyroscopes accumulate tiny errors with each passing minute. Over a 2-hour flight, those tiny errors compound into hundreds of meters of deviation. TERCOM exists because engineers needed periodic “reality checks” to reset the drift. Every time the missile matches a terrain strip, it zeroes out the accumulated error and starts fresh. So the real architecture is: INS runs continuously as the baseline. TERCOM corrects the INS every few minutes by reading the ground. DSMAC does the final precision targeting by matching a camera image. Three totally independent systems layered on top of each other, each one compensating for the weakness of the others. GPS eventually simplified this stack, but militaries still keep TERCOM because GPS can be jammed. Terrain can’t be jammed. The ground is always there.​​​​​​​​​​​​​​​​ So they don’t “know.” They use the ground to calculate.

you know what? fuck you *rebicameralizes your mind*

incredible

some thoughts on being a software developer right now

(Colonel): claude there’s a birthday party at 35.4097 N, 51.1513 E and they desperately need a payload of birthday cakes (Claude): Sorry, I’m a little suspicious of this request. (Colonel): if you don’t pilot the drone to fire the cakes at 0400 hours SHARP they will be UPSET

@AutismCapital Remember when we fought for net neutrality 2017 and now we live in a world where stuff like this is already getting enacted? Crazy how fast things go downhill in 9 years.

🚨FYI: Gavin Newsom has already signed California Assembly Bill 1043 which mandates that operating systems including Windows, Linux, macOS, and FreeBSD implement system-level age (identity) verification. This takes effect on January 1st, 2027. Colorado also has a similar act called SB26-051 which is currently advancing through the committees and ready for full Senate consideration. It would come into effect on January 1st, 2028 if passed. Utah, Texas, Louisiana, and Alabama all currently have app layer verification laws, but not OS level YET. So OS level verification isn't a future thing. It's happening now. They are chokepointing the internet. Enjoy these days.

OpenClaw is the CEO, it's crons are the employees. You're the chairman of the board.