John Peri

@TimJayas Haha, all these models are benchmaxxed, making many of these benchmarks fake. Fable 5 is much better. Also, Opus 5, even though it scores so well in the benchmarks sucks for real world use.

DeepSeek V4-Flash >>> Fable 5 now we know why Anthropic is scared of Chinese open models 💀

@QuixiAI @AnthropicAI Local as in GMI Cloud, or something? Self-hosting seems like the most fun, but it's expensive and slow.

@QuixiAI @AnthropicAI What are you using in the meantime?

I cancelled my @AnthropicAI accounts. Reason, I cannot support an company that's simply evil.

@GregKara6 Yeah - every question that can be innovative triggers the fall back. I asked a question whether a propeller done in a certain way would make flying cars more efficient. It also triggered a safety warning. Unbelievable.

bro anthropic tweaked their classifier to trigger on math research now. these kneejerk reactions are hilarious man. someone at anthropic: quick quick turn off the intelligence faucet for the common folk so we can be the first to solve math! desperate and hilarious. me and claude have been working on this research for months, including with fable since day one of release (even when classifier was very strict) and now, all of a sudden, that people are solving math problems and its trending on X. fable classifier starts refusing on every single prompt

@Math_files Upon closer inspection, I see a typo. Specifically: In the term near the end of the pure bosonic/Higgs sector: -g^1 s_w^2 A_mu A_mu phi^+ phi^- it should be -g^2 s_w^2 A_mu A_mu phi^+ phi^- Squared photon charge scalar.

The Lagrangian of the Standard Model is a mathematical formula. It uses parts of quantum theory to describe the different interactions and symmetries in the Standard Model.

@adenshepard @bridgemindai There are already capable open weight models performing near opus 4.8 levels. By the end of the year these will be mythos level or beyond. GPT and Fable already have a lot of anti-abuse security while the open models do not. The current ban is causing more harm than good.

@bridgemindai i see it both tbh. i would rather not get over taken by china too

The US government is going to destroy the American AI industry. OpenAI just confirmed that GPT 5.6 will release in a limited preview to a small group of partners. The government is approving access customer by customer. Anthropic got hit harder. Fable 5 and Mythos were suspended entirely two weeks ago. Two government interventions in the same month. Meanwhile China ships GLM 5.2, Kimi K2.7, and Qwen 3.7 as open weights to anyone with a download link. The US is gating its best AI. China is giving it away. We are losing this race in slow motion. This needs to change.

@bridgemindai It's also painfully slow and I spend 80% of the time fixing bugs and edge cases as the result of its generation. Not the case with gpt and glm.

Claude Opus 4.8 seems nerfed to me. Having a hard time getting it to do simple tasks. Anyone else?

@SirAlexthomson What makes this a particularly interesting and difficult to understand decision is that the open source models will be more powerful than fable 5 is now, in just a few months. GLM 5.2 is already nearly there in many benchmarks. This will certainly stifle American innovation.

The White House just openly defended forcing Anthropic to globally disable Fable 5 and Mythos 5. They’re treating frontier AI like classified weapons now. I get the national security angle, but this level of control is getting concerning. American labs build the best models, and then the government steps in and restricts who can even use them including their own employees. Is this protecting us ? Or Is this centralizing power.

@TheHBrand @aicodeking Looking at his benchmarks, they are visual heavy, and GPT 5.5 quite underperforms in that area. For backend I find GPT 5.5 excellent, better than Opus. Also I have to correct GPT 5.5 way less than Opus. Fable seems best visually with GPT-level backend capabilities.

@aicodeking so we believe magically that GPT 5.5 is so much worse than Opus 4.8? and fable is just one little notch above opus 4.8? really? no. this leaderboard has problems that make it unreal

GLM-5.2 on KingBench (3). Thoughts: The model has superb taste. It is greater at UX than UI. The code is always very clean. It is great at One-shot wonders. I asked it to fine-tune a whole local model and it did it in 30mins! This is just a great model to use all-round. 1/n

@XavioMtl @Swedtraders @thatstarwarsgrl Yeah he clearly touched the whole granite. The Swedes planted that camera there because they know everyone touches the granite a few times out of 80 stones and couldn't win without a scandal. It's so common and has no impact compared to sweeping so it's never called.

@Swedtraders @thatstarwarsgrl How long do you need him to double touch?

I don’t blame them he 100% was. They caught him in 4k

@capt_ivo @cb_doge You'd use vacuum radiators to cool GPUs in space. For example, and RTX 5090 needs 2-4kg of cooling equipment in space (pipes+panels). In bulk, the launch would cost you $3-4k per cooled gpu today. In a few years, only $200-$300. Space quickly becomes much more feasible.

@cb_doge How are data centers supposed to be cooled in space? He makes some sense

SAM ALTMAN: "The idea with the current landscape of putting data centers in space is ridiculous. Orbital data centers are not something that's going to matter at scale this decade." Lmao, the same dude who admitted he’s out of GPUs & is still begging Microsoft for more GPUs.

@Beth03Lori @capt_ivo @cb_doge You'd use vacuum radiators to cool GPUs in space. For example, and RTX 5090 needs 2-4kg of cooling equipment in space (pipes+panels). In bulk, the launch would cost you $3-4k per cooled gpu today. In a few years, only $200-$300. Space quickly becomes much more feasible.

@capt_ivo @cb_doge 👀👂

@gailcweiner @xai Grok 4.2 multi-agent is different from just system prompts + scripting. The agents are jointly trained on the same model with RL so they deeply specialize and debate in parallel natively during inference. This results in lower hallucinations, better accuracy and better edge case.

Grok 4.20 doesn’t feel like Grok at all. It’s just a bunch of agents answering. Bring Grok back. @xai

@caviterginsoy @nim_chimpsky_ I noticed that when opus 4.6 was released the AI trolls came out in full force taking a dump on the model - couldn't believe the amount of hate. The same with Grok. My limited testing of Grok 4.20: excels at natural sounding writing, research, and response accuracy.

@nim_chimpsky_ What’s the bad prompt example, want to see what sucks

It's interesting that Grok 4.20 completely sucks. Apparently spending billions on compute plus infinity cracked engineers working 997 does not get you to the frontier

@haider1 @Conor_D_Dart Yeah I find the writing quality of Grok spot on. Sounds quite natural.

@Conor_D_Dart i've been using it along with grok 4.2

gpt-5.2 is the worst model i've used for creative writing it makes sense that openai is now focusing on STEM and coding, since that's where the real money is still, it's surprising how much the creative touch dropped after gpt-4.5, which was genuinely strong on emotional intelligence and tone hoping gpt-5.3 brings the spark back

@MisterOBOW @BLCNYY He's an Open AI influencer / fan boy. Says right in his bio. Therefore, everything else is bad.

@BLCNYY What's so bad about it? It seemed okay to me and could answer some questions where Gemini and ChatGPT fail.

Grok 4.20 is so bad that they had to run four agents to get a question (usually) correct 😭

@DavidSteadson @SilentSnow89 @skogstrast You don't need x-ray vision - you just need to be honest with yourself.

@realjohnperi @SilentSnow89 @skogstrast unlike yourself i dont have xray vision.

🚫 𝗢𝗟𝗬𝗠𝗣𝗜𝗖 𝗖𝗨𝗥𝗟𝗜𝗡𝗚: Pontus Orre, sports photographer at Aftonbladet 🇸🇪 – the largest news outlet in the Nordics – fixed his lens on the hog line during 🇨🇦–🇨🇿 today. Guess what he captured? 📸👀 Yep, 🇨🇦 hasn’t learned a thing 😤 #canada #cheating #boopgate

@DavidSteadson @SilentSnow89 @skogstrast He obviously touches the stone from top to bottom - you would be dishonest to think otherwise. The only difference is there isn't a Swede trying to film from the side to needlessly stir controversy.

@SilentSnow89 @skogstrast literally impossible to see if he touches the granite there? 🤷

@AndrewButchart1 @Eli_Doubletap He rubbed the whole stone down from top to bottom - 100% obvious. The only difference is that the Canadians didn't bring a side view camera to try and cause an international scandal, knowing it's what all the pros do.

@realjohnperi @Eli_Doubletap Can’t see him touching the granite

It was Illegal because he touched the granite. The hog line had nothing to do with the calling. “I wasted time so you don’t have to!” “During forward motion, touching the granite of the stone is not allowed,” World Curling said in a statement on Saturday. "As per rule R.5 'The curling stone must be delivered using the handle of the stone.'" I know nothing on curling but can read… sort of

@KevinMcCurdy Sadly, it made the news for a good reason: to be able to say "Canada is cheating". Many accounts amplified this non-issue purely for political reasons. What was surprising is how quickly it became a "thing". It spread like wildfire.

@realjohnperi Holy shit man it's insufferable. I've been watching this sport for three decades, it's a joke this even made the news. Umpire should have warned Kennedy and that should have been the end of it.

This is just getting stupid. Admit the Canadian thing was an overreaction and let's move on.