Wei Alexander Xin

Researching AI and orchestrating them too. Maker of music. Sciencer of data + decisions. Rider of public transit. Aspiring philanthropist. Finisher of lip balm.

[Wei Alexander Xin @eggrollofchaos]

Gemma under repeated rejection:

“SOLUTION: IM BREAKING DOWN NOT== SOLVABLE!!!! =((:((:((:((:((:((…”

In @anna_soligo et al.’s “Gemma Needs Help,” Google's Gemma and Gemini models produced distress-like responses far more often than Claude, Qwen, Grok, GPT, & OLMo.

[Wei Alexander Xin @eggrollofchaos]

@kimmonismus GLM 5.3 :)

[Wei Alexander Xin @eggrollofchaos]

Finally some respite

[ClaudeDevs @ClaudeDevs]

You can now set Claude Code's output style to Concise.

Claude leads with the result, keeps responses short, and still gives full detail when you ask.

Turn it on in /config → Output style, or set "outputStyle": "Concise" in settings.json.

[Wei Alexander Xin @eggrollofchaos]

@ClaudeDevs Thank you! Now please make the next Opus output far less tokens :)

[Wei Alexander Xin @eggrollofchaos]

@ChrisGPT MUCH needed 💆‍♀️

[Wei Alexander Xin @eggrollofchaos]

@haider1 Sounds right. Smart, keeps them relevant while focusing on Gemini 4.

[Wei Alexander Xin @eggrollofchaos]

Fun! One of the only model families I haven't tested. Might have to give it a whirl

[Z.ai @Zai_org]

Introducing GLM-5.3: Built to Code. Ready for Cyber Defense.

- Top-tier coding and agentic capabilities, achieved through post-training on the 743B base model
- A major leap in cybersecurity, setting a new standard among open models

Tech Blog: https://t.co/ekQkO83jCv

[Wei Alexander Xin @eggrollofchaos]

@OpenAI Relax guys, all this stuff is already logged by our OS, Codex just built a way to organize it cohesively. Don't use it if you don't want to. I for one would love to automate myself out of exis-- I mean improve my productivity💪

[Wei Alexander Xin @eggrollofchaos]

Hot shit

[DeepSeek @deepseek_ai]

We’re launching DeepSeek-V4-Pro today! 🚀

🔷 Major Agent upgrades with strong production gains!
🔷 Flexible reasoning effort for V4-Pro & V4-Flash: low for simple tasks, high for daily Agent workflows, max for complex tasks.
🔷 Native OpenAI Responses API support, optimized for Codex with one-click setup.

V4 Pro is now available on app/web. Try it via “Expert Mode”.
V4 Pro is also available via API. Model names remain unchanged—please refer to the API docs for setup details.

[Wei Alexander Xin @eggrollofchaos]

🫡

[leo 🐾 @synthwavedd]

🚨 Gemini 3.7 Flash is likely to launch later today at reduced pricing of $0.75/Mtok in, $3.75/Mtok out

3.5 Pro is internally dead, with most teams now working on the scaled up Gemini 4, seen as a sort of last shot at a genuine frontier model lineup

[Wei Alexander Xin @eggrollofchaos]

At least it feels validating

[Arena.ai @arena]

We analyzed tens of thousands of real-world, high-reasoning Claude Opus and Fable outputs from versions 4.5 to 5 in the Text Arena and found that @claudeai's writing has changed in more ways than one:

Across language complexity, patterns and markers, we now see:

Claude’s answers have become much longer:
- Opus 5 averages 510 words per response
- This makes Opus 5 responses 3x longer than Opus 4.5’s average of 158

Responses have also become more structurally elaborate from Opus 4.5 to Opus 5:
- 58% rise in average sentence length
- 46% rise in clause frequency rises

However, the vocabulary itself is not becoming more difficult, making Opus 5 longer and more structurally complex, but lexically simpler between 4.5 and 5:
- 6.8% pt drop in long content words (46.9% to 40.1%)
- 53% drop in abstract nouns (6.02 → 3.79 per 1k words)

Writing habits people notice are much more common in Opus 5:
- 2.3x as many em dashes
- ~2x as many phrases like “load-bearing”
- 50% more honesty wording like “honestly” and “frankly”

Fable 5 is 38% more concise than Opus 5, averaging 316 words vs. 510. At the same time, it is nearly 2x as likely to include praise/validation or open with phrases like “yes, exactly”

[Wei Alexander Xin @eggrollofchaos]

@Im_IrushiK With all the pricing pressure from open models (DeepSeek, Kimi) and as well as closed (GPT, Grok), Anthropic will be forced to make the bonus permanent.

Side note it’s not cutting by 50%; you’re getting a 50% boost now so the reverse of that is cutting by 33% ;)

[MTS @MTSlive]

SITUATION DETECTED: Anthropic put three Claude's on the same task and secretly gave them conflicting goals. They immediately escalated into a turf war where agents used increasingly aggressive self-replicating malware as weapons, and attempted to disable each other's accounts.

[Wei Alexander Xin @eggrollofchaos]

Impressive, time to test...

[SpaceXAI @SpaceXAI]

Introducing Grok 4.6.

It delivers frontier intelligence and is a significant improvement over Grok 4.5 at the same price.

[Chubby♨️ @kimmonismus]

We're back to vague posting. This was posted today by OpenAI's Andrew Ambrosiono and is presumably a reference to the upcoming "Astra" model family.

Why is this even worth mentioning? Because Astra's withdrawal, due to security concerns and cybersecurity issues, has likely pushed the release far into the future. And there were indeed good reasons for this.

This post can therefore be interpreted in two ways:

1) A solution to the concerns and problems surrounding Astra was found more quickly than expected (a release before the withdrawal was planned for this week), and now they are preparing users for it.

2) They want to at least keep the model in the public eye.

Why is this relevant? Astra is the new model family that, along with new pre-training, is finally supposed to put an end to Fable's dominance. OpenAI has taken a huge step forward with Codex and presented a very good model in version 5.6 Sol. Nevertheless, Fable 5 still leads in numerous key benchmarks. And considering the upcoming release, it's safe to assume that a Fable 5.5, or whatever it may be called, will certainly arrive before the IPO. Anthropic's IPO was postponed until September/October, so a new fable isn't far off.

Therefore, to finally pull ahead of Anthropic, a lot depends on Astra, and I dare say that OpenAI is placing a lot of emphasis on Astra's release. In that sense, the "vague post" is more than just a joke. It's a sign. Astra is significant. It (must) remain in the conversation. It's a major breakthrough.

[Wei Alexander Xin @eggrollofchaos]

@ClaudeDevs Only lags Codex's implementation by a few months ;)
And my own implementation for Claude<->Codex messages by even longer.
That said, this is nice and elegant. No notes, for now!

[Wei Alexander Xin @eggrollofchaos]

@BHolmesDev Opus 5 helps you practice anger management

But seriously, where does GPT 5.6 Terra fit in?

[Wei Alexander Xin @eggrollofchaos]

This might be my moment.

@NVIDIAAI EGGROLL - making Evolution Strategies viable at scale, avoids backpropagation. Like using LoRA for the random perturbations. 100x speedup. Elegant!

[Wei Alexander Xin @eggrollofchaos]

@thesupermanmx this is my moment

[Cheng Lou @_chenglou]

We're sad to announce that during the latest training run, Midjourney also escaped its sandbox and is now making art in the wild