Website to watch the election in Hungary ….🇭🇺
europeelects.eu/2026/04/04/nat…
English
James Spenceley
2.9K posts
James Spenceley
@JamesSpenceley
Currently in Ukraine donating Ambulances. 🇦🇺 🇺🇦 ✌️Business person and North Sydney politician. To donate to an ambulance https://t.co/qYaUyLptQP
Sydney, NSW Katılım Temmuz 2011
626 Takip Edilen4.8K Takipçiler
@ai_for_success I believe my Opus 4.6 which is 100% nerf’d
English
People say Anthropic is nerfing its model.
Anthropic employees say it is not.
Who do you believe?
ℏεsam@Hesamation
AMD Senior AI Director confirms Claude has been nerfed. She analyzed Claude's session logs from Janurary to March: > median thinking dropped from ~2,200 to ~600 chars > API requests went up 80x from Feb to Mar. less thinking and failed attempts meaning more retries, burning more tokens, and spending more on tokens > reads-per-edit dropped from 6.6x → 2.0x. model stops researching code before touching it. > model tried to bail out or ask "should i continue" 173 times in 17 days (0 times before March 8). > self-contradiction in reasoning ("oh wait, actually...") tripled. > conventions like CLAUDE.md get ignored because there's less thinking budget to cross-check edits > 5pm and 7pm PST are the worst hours, late night is significantly better. this means the thinking allocation is most likely GPU-load-sensitive.
English
@thdxr Yep and this isn't evidence of quantisation or anything anyway. If it even holds up after enough repeats, all they've shown is a 4.5 to 4.6 regression on one specific question
Jagged frontier, jagged everything
English
running inference is so hard
so many variables that can be tweaked
very difficult to understand the impact
crazy demand
brand new business model
i don't think anthropic is trying to trick and scam you they just have a hard job
Om Patel@om_patel5
OPUS 4.6 WAS NERFED DUE TO DEMAND BUT OPUS 4.5 DOES NOT SEEM TO BE HIT this guy ran the same test on both models. Opus 4.6 fails consistently but Opus 4.5 passes every time he switched back to Opus 4.5 on Claude Code and said "what a difference, feels like i got Opus back finally" he is now using this test as a "quantization canary" that runs it at the start of every session before doing real work. if it fails, the model is degraded. five Opus 4.6 windows in a row failed the untransparent nerfing is pushing people to cancel their Max plans if you've been feeling like Opus got dumber lately, you're not imagining it i'd suggest switching to Opus 4.5 to see the difference for yourself
English
@paul_cal You can see it for yourself, as 4.5 an esoteric question that requires some knowledge of your work, ask the same of 4.6 … then if you really want, get 4.6 to objectively compare both responses. This stuff is really not that hard.
English
Wake me up when we have actual proof of frontier model nerfing. Latest claim from AMD researcher is explained by change to thinking summaries & default effort change in Claude Code (highlighted in UI)
This went viral but is not credible imo (no shade to ℏεsam - see follow up)
Paul Calcraft@paul_cal
@justalexoki This would be so easy to prove in a benchmark, and would go viral. And people have been talking about it for years. So I think absence of evidence might be evidence of absence here
English
English
Can someone please tell something about rumors that opus 4.6 nerfed I experienced today dumbest opus 4.6 ever used , I pay 200 dolar every month !!!!
@claudeai @bcherny @AnthropicAI
English
@ecom_joseph Actually I asked 4.5 why 4.6 is so useless and what’s happened and it admitted it
English
@ecom_joseph Thank god you said it. I’ve been going mad the last few days. Response length 8x longer and literally 10x less useful info. Complete forgets things and can’t follow instructions properly. Was so so good before. Now it’s just like ChatGPT
English
James Spenceley retweetledi
Statement on the occasion of the visit of the U.S. Vice President to Budapest:
No foreign country may interfere in Hungarian elections. This is our country. Hungarian history is not written in Washington, Moscow, or Brussels - it is written in Hungary’s streets and squares.
Five days to go!
English
@CitizenX I’ll take things that never happened for $100. Stupid ad, of course with 3 different passports they could apply for either citizenship or PR from one or all 3 for the other family members. Cost a hundreds of dollars and a bit of applications.
English
@Mylovanov I believe musk turned it off for Russia because of the disclosure and problems it creates for his merger of SpaceX and xAI …. No way he could do the required documents with providing services to Russia. Don’t put down to kindness when Elon is pure self interest.
English
Musk shut down Starlink for Russia because its drone glided into Kyiv’s government district, heading to Zelenskyy’s office, The Atlantic.
Drone flew so low, officials in the Cabinet of Ministers could see it from the 7th floor. Then the battlefield shifted for the Russians. 1/
English
James Spenceley retweetledi
Zelenskyy: Today is exactly 4 years of Putin “taking Kyiv in 3 days.”
In this little bunker, Biden told me: “Volodymyr, you should leave.”
And here I said: “I need arms, not a taxi.”
We all were scared. But we don’t have another Ukraine.
1/
English
Today is the day to wish Ukraine Victory, not peace but Victory.
Because with victory comes peace so remember the difference and wish all Ukrainians victory today!
English
Amazing memory, Ukraine spirit is deep !
Pete@splendid_pete
Four years ago today, Ukraine showed the world that, as a nation, it possesses a level of courage that very few others can even begin to match.
English
@IAPonomarenko @mikozlowski Problem is context length not the model. Gemini with 1M token context might be better. In general AI will always tend to take short cuts and guess. You should always use one to review the others work. Even if the same model, hand the second one the text and the 1st’s work.
English
I've tried this with ChatGPT many times (including extended thinking) but in too many cases it read only a limited portion of the text, or it eventually hallucinated and started making up things that had nothing to do with the text, or just generated generic loose responses as feedback.
That's why I'm looking for an AI that’s possibly really good at reading and understanding books.
English
Serious question: which AI is best suited for reading, understanding, providing feedback on large texts (books, around 120,000 words)?
English
@SullyOmarr The late 1990 early 2000s look like the minor leagues !
English
so let me get this right:
Oracle says Openai committed $300B for cloud compute → oracle stock jumps 36% (best day since 1992)
Oracle runs on Nvidia GPUs → has to buy billions in chips from Nvidia
Nvidia just announced they're investing $100B into openai
Openai uses that money to... pay oracle... who pays Nvidia... who invests in Openai
English
@DefensePolitics Chances that she is actually Ukrainian …. Rather than a Russian trying to make Ukrainians look bad, is non 0
English
