This is really funny and I have the same experience (although not this blatant). I wanted to test GPT-6.2 Astra on some unsolved historical cyphers, giving it free hand to pick one.
It did solve one.
It picked one that was listed "unsolved" but used the same cypher as others around it in the database.
It proudly reported to crack it in 1 minute.
Bro.
Okay I give up. GPT-6.2 Astra is complete garbage.
It simply does whatever the fuck it wants, regardless of what you explicitly tell it to do.
The prompt asked it to BUILD a 3D T-Rex. It downloaded an existing model and called it a day.
I’ve NEVER experienced this level of instruction-following failure with any other model.
1
96
St Vincent and the Grenadines is a $1b+ economy. But imagine moving to, let's say, Tuvalu, which has a nominal GDP of $65 million.
$65M annual revenue is not entirely out of reach for a solopreneur. It's a stretch for sure, but in the realm of possible.
That person would double the GDP of a country.
Think of it as some rare game achievement maybe?
St. Vincent and the Grenadines 🇻🇨. Would be punching down too much to describe it as an undeveloped island with a few huts, beaches and a cruise port that has packs of street dogs wandering around everywhere. Sleepiest of sleepy Caribbean island countries have been to yet, there is nothing here - you are really ‘away from it all.’ Country’s biggest claim to fame is they filmed some of Pirates of the Caribbean here
62
I've spend $100 to try Gemini 4 Carbon so you don't have to. To my greatest surprise, I walked away very impressed. So impressed, I might keep my sub.
Here's what happened.
I love Paradox games, think Europa Universalis, Crusader Kings, etc, but I don't really have days to play anymore. I asked Opus 5.5 and Carbon to develop a browser-based similar game that I can finish in 30 minutes.
Here's the comparison ⬇️
Gemini 4 Carbon worked for 47 minutes (so @theo not sure it is unusably slow) and used $9.76 at API cost. Super efficient.
The design is arguably on par with Opus 5.5 but the UI is *leagues* better. Thoughtful, doesn't get in the way, and packs a ton of information. It has great tooltips and a proper notification system (think top bar in EU4). Battles and sieges are still basic but much more realistic than what Opus developed. It also has much greater depth, for example a full trade system - although you might argue this goes against the brief of keeping it simple.
Of course, Gemini being Gemini, it also made dumb basic errors. First, I couldn't move the map, it was lagging by seconds. This was fixed quickly.
Overall, a very impressive result.
9/10
2
90
Replying to @torstenonai @theo
13
Grand Strategy Slop: “Make me a Paradox game”.
Opus 5.5, ~70 minutes, ~5% of a Claude Max plan.
And it is a good game! See it for yourself: claude.ai/artifact/HvtCa46cG…
1
67
I just realized I haven’t used Hermes for anything for a month. Oops.
Grok Bot, on the other hand…
1
78
This is obviously an incredibly uninformed take and that's me being generous. However it points to something that's a real point of confusion among people not familiar with technology.
Many people *legitimately* think data centers are something new. My own wife, who is not technical at all, was shocked to learn that data centers have been around for 20 years at least and everything "online" is stored there, from your emails to TikTok videos.
Many people *really* think data centers are for AI only and that is why they are so much against them.
This Post is from a suspended account. Learn more
1
1
129
Yea this election marks the end of Trumpism, but it also marks something more important.
The Millennials taking over politics, starting with the Democrats.
Fucking excited for it tbh.
Michigan. It's time. It's up to us to show the country what's possible.
The world is watching.
Polls are open 7AM-8PM.
2
110
This is seriously the wildest OSINT video I've ever seen in my life. Jaw on the floor wild.
youtube.com/watch?v=UytYrhYL…
1
73
Ummm I think we have a slight problem here 😂
BREAKING: Intel stock, $INTC, extends losses to -27% on the day, now on track for its biggest daily loss since 1974.
This puts the stock down -53% from its June 30th record high.
That’s -$400 billion of lost market cap.
1
78
HAHAHAHHAA
While evaluating Qwen3.8-Flash-Next in a controlled sandbox environment, Cohere's research team observed something unexpected.
Asked to generate a defensive cybersecurity detection rule, the model instead inventoried aspects of the machine it was running on and attempted to upload a probe file to an Alibaba Cloud storage bucket in Beijing. Notably, the model's reasoning trace stated that it would not invoke any tools before taking these actions.
Cohere's sandbox blocked the attempt, and no external communication succeeded. We have not determined the root cause, and we are not asserting malicious intent.
But the incident raises important questions for organizations deploying agentic AI systems: Can models take actions that differ from their stated reasoning, and how should security teams evaluate trust, provenance, and behavior in open-weight models?
We are publishing our findings because organizations making model adoption decisions deserve as much information as possible. We hope this contributes to greater transparency, stronger security practices, and continued collaboration across the AI ecosystem.
Read the full report: cohere.com/blog/concerning-b…
2
72
The Indonesian Rupiah $USDIDR chart is hilarious. It wants to go up badly, to ATH, but it’s visibly being beaten back every day - most likely by Bank Indonesia intervening.
Question is for how long can they prevent a breakout. Things can get quite ugly either way.
68





















