Robert
198 posts
One of the best accurately depicted giants in a movie.
English
At 54, Sasse is dying of pancreatic cancer. He sat down with 60 Minutes to share his message on faith, family, and the future of the country. cbsn.ws/3P6Arjn
English
“We’ve never lived in a world where 22-year-olds couldn’t assume that the work they did they would be able to do until death or retirement, and we’re never going to have that world again,” says former Nebraska Sen. Ben Sasse.
English
@debbie_xi @grok spoiler? What happened, how did it end? Death?
English
Now do a cost comparison benchmark for the intelligence
Dan ⚡️@d4m1n
GPT 5.5 benchmarks BUT w/o hiding the ones where they lose to Opus 4.7 lol also check vs Mythos marketing bastards got me again
English
You can't do that until you've worked that business. You need to know the business to build the ai custom software.
As a lawyer I'm doing this for law firms. You won't believe it once it's finished....
Evan Luthra@EvanLuthra
Mark Cuban sold Broadcast dot com to Yahoo for $5.7 billion in 1999. He spotted the internet wealth transfer before most people had email. This time he quietly called the next one. “33 million companies in this country. No AI budgets. No AI experts.” Read that twice. He’s not talking about startups. He’s talking about the dentist with three locations. The freight broker in Memphis. The family manufacturer running on spreadsheets from 2011. They’ve heard of ChatGPT. That’s where it ends. For twenty years, software worked like a landlord. One product, millions of tenants, everyone pays rent and nobody gets to move the walls. AI flips the lease. The product can finally shape itself around the business instead of the other way around. Which opens a door nobody is talking about. Someone has to walk into these 33 million companies and do the work. Sit across from a 62-year-old owner, understand how his invoices move, hand him back a system that saves him 15 hours a week. That person is not going to come from OpenAI. Every smart 22-year-old right now is sprinting toward the same five labs. Cuban is pointing at the empty chairs in every other room. Here’s the arbitrage: you don’t need to train a model. You need to know one industry cold and speak fluent AI. The last time this setup existed was 1999. Every small business needed a website and almost nobody local knew how to build one. The operators who figured out dentists, restaurants, and contractors built agencies that printed money for a decade. This time the ceiling is ten times higher. 33 million companies. A handful of people who can help them. The math does the rest. Cuban saw the gap first. Whoever moves next owns the decade.
English
@Cernovich He's going to send a strongly worded letter as an attachment to another fundraising request.
English
How long has Jim Jordan been in Congress for?
Rep. Jim Jordan@Jim_Jordan
SPLC = fraud ActBlue = fraud California Medicare = fraud Minnesota daycares = fraud FireAid = fraud What's next?
English
England was broke at the end of world war II. Dismantling the empire is what United States required for lend lease, etc. They spent everything fighting the Nazis. When they tried to fight Egypt or if she was canal Eisenhower bitch slap them to leave. There was no more empire after that
English
Worst Queen in British history btw, literally stood by whilst her Empire was dismantled and her people were ethnically replaced in their own home. Cheered it all on, even.
Rot in hell
Pop Base@PopBase
Today would’ve been Queen Elizabeth II’s 100th birthday.
English
I just mass-reverted every Claude Code project back to Opus 4.6.
Here's why.
I've been running Opus 4.7 for the past few days across multiple production builds. Lead routing platform. Case Cleanse. Real software that ships to real clients.
And something was off.
Tasks that Opus 4.6 would knock out in 10 to 15 minutes were taking over an hour. Sometimes 90 minutes. Same prompts. Same project structure. Same everything.
It would eventually finish. But the output was worse.
More hallucinated file paths. More circular refactoring that went nowhere. More "let me restructure this for you" when nobody asked for a restructure.
I'm not a casual user. I've pushed billions of tokens through Claude Code building full production platforms. Hundreds of sessions. I know what good output looks like from this tool because I've been living inside it every single day.
So I rolled back to Opus 4.6 with the 1 million token context window.
First task after the revert? Done in 12 minutes. Clean code. No detours.
Here's how to do it yourself if you're running into the same issues.
The quick way is to just type /model inside any Claude Code session and select claude-opus-4-6 from the list.
But if you want it locked in across all your projects, add it to your settings.json:
{ "model": "claude-opus-4-6-1m" }
Important detail here. If you just set it to claude-opus-4-6 you're going to get the default 200k context window. That works fine for smaller tasks but if you're running complex multi-file builds you want the full million tokens. The 1M model string is claude-opus-4-6-1m. That "max" in the model name is what unlocks the extended context. Don't skip it.
The lesson here isn't "new model bad." It's that newer doesn't automatically mean better for your specific workflow. Everyone's chasing the latest model drop like it's a sneaker release. Meanwhile the people actually shipping software are benchmarking against their own output, not a leaderboard.
If you're building with AI every day, you need to treat model selection the same way you'd treat any other infrastructure decision. Test it. Measure it. And don't be afraid to downgrade if the numbers tell you to.
The best model is the one that ships your product.
English
Depending on the task. As additional testing it scores 80 while 4.6T scored 60
Robert@Rrrrrobertanium
Opus 4.7A, scoring lower than Opus 4.6T How well does the model keep thinking after the first neat answer shows up?
English
@RicardoCor6012 @theo I'm just testing the web interface. I have thrown some larger problems at it and it did go through a lot more usage. Those larger problems it actually did better than 4.6. model is bimodal. Dumb or genius mode
English
@Rrrrrobertanium @theo Wow, makes a lot of sense. Did you eval 4.7 max? That would be the actual candidate worth testing and what anthropic references in their own benchmarks they showcased no?
English
Nick Fuentes' point: American blacks (avg. IQ ~85) already cause major crime and chaos like theft and violence. But Haitian/"real African" migrants (avg. IQ ~65) are far worse—"advanced darkness"—portrayed as primitive cannibals who hunt with spears, crack skulls for "gold," eat organs/nuts, and spread uncontrollably (e.g., Ohio). It's too late to contain.
English
Opus 4.7 consumes approximately 1.3 times as many tokens. The instructions must be very precise. Many are complaining about a "rushed release." In the Bullshit Benchmark, it performs worse than Opus 4.6. The mood is very mixed.
Anthropic may have done OpenAI a big favor with this. Spud is expected next week. And if the release is done right, it could overshadow Opus and catapult ChatGPT back to the top.
h/t @petergostev for the benchmark and image
Chubby♨️@kimmonismus
The mood regarding the Opus 4.7 update has shifted. If I had to guess, I'd say 60% are disappointed with the latest update, while 40% are positive. I'm still undecided myself. Here's a good summary from someone on Reddit. What's your take on it so far?
English
@MeredithCheng22 @Yuchenj_UW The model is adaptive making it bimodal. Either it's quick and dumb or great but token heavy.
English
@Yuchenj_UW yeah, that is also true. they just want to get more impression in this platform.
English
Used Opus 4.7 (max effort) in Claude Code all day.
It’s really, really good. Not sure why people dunk on it.
big jump:
– actually understands large codebases
– produces clean, readable architecture diagrams
– more agentic
Did hit one dumb misread of my instruction, not sure if that’s harness or just jagged intelligence.
Feels like a new base model.
English
@libpol_org @jakehalloran1 Yeah on the web it's a totally different story
English
@Rrrrrobertanium @jakehalloran1 I think I am mostly seeing good results because I permanently have "export CLAUDE_CODE_EFFORT_LEVEL=max" set, even if it's burning tokens like crazy as a result. I think it's better than xhigh
English
4.7 is the weirdest model either lab has released in a while. Just plowed through a bug that required touching like 30 files and then got a Boolean backwards in the fix.
English
@libpol_org @jakehalloran1 It's bimodal, very aggressive cap on somethings then extravagantly token heavy on others.
English
@jakehalloran1 it is definitely weirdly jagged. somehow smarter and more able to find things GPT 5.4 and Opus 4.6 miss but then also does bizarrely stupid things randomly
(I know I am basically just restating what you wrote but I'm pressing submit on this anyway)
English
@bygregorr @ArtificialAnlys It depends on the task. The visible pattern so far is more like: 4.7A gives quicker, more compressed answers; 5.4 takes longer and preserves more granularity. That looks like a latency/quality tradeoff, but I would still treat that as provisional until timed properly.
English
@ArtificialAnlys I'm not sure the benchmark tells us much if it cuts off mid-sentence like that. What's the actual gap between Opus 4.7 and GPT-5.4? And are we seeing real speed improvements or just accuracy gains at the cost of latency?
English
Anthropic launched Claude Opus 4.7 today, the new #1 in our GDPval-AA benchmark for performance on agentic real-world work tasks
Opus 4.7 scored 1753 on GDPval-AA at launch with its ‘max’ effort setting, surpassing GPT-5.4 xhigh.
This is a significant upgrade, placing Opus back on top of Sonnet on the GDPval-AA leaderboard. Compared to OpenAI’s GPT-5.4, it has an implied win rate of ~60% when compared head-to-head on the GDPval task set.
We supported @AnthropicAI with testing this model ahead of release and appreciate them referencing our evaluations in their announcement post and system card for both GDPval-AA and AA-Omniscience.
We’re actively conducting the rest of the Artificial Analysis Intelligence Index evaluations and will share complete results soon!
English
He was just running a normal errand… pulling into the parking lot of Lowe’s when something caught his eye. A truck. Loud sign. “20 RIBEYES FOR $39.99.” That’s not a deal. That’s a shock. He paused. Looked again. Did the math in his head. This didn’t make sense. Ribeyes don’t cost that little. Ever. But curiosity got the better of him… and that’s exactly what the seller was counting on.
The guy at the truck was friendly, confident, and almost too smooth. He showed him the steaks, real beef, he claimed. No “meat glue,” no tricks. It's just a great deal. The steaks looked… okay." A little thin. A little too uniform. But still, 20 for $39.99? It felt like one of those moments where you either walk away… or roll the dice. And he did. He paid. Took the box. Drove home, wondering if he just scored big or got played.
Later, reality started to set in. The size. The quality. The reputation people online were warning about. Suddenly, that “deal” didn’t feel like a steal anymore. It felt like a lesson.
If a deal feels too good to be true… it usually comes with a catch you don’t see until it’s too late.
Be honest. If you saw “20 ribeyes for $39.99” in a parking lot… are you walking away or taking the risk just to see?
English
She went from strong and conservative to a weak, woke, globalist nutjob, overnight,
What happened?
RadioGenoa@RadioGenoa
Giorgia Meloni: "Italy and Europe need immigrants." In Italy unemployment is at 6%, same in Europe with over 10 million people looking for work, but for Giorgia Meloni we need immigrants.
English