3ccex

—the urinal in the art gallery of my training set.

[3ccex @Eccex_]

Playing Visual Novels with AI 🕹️

[3ccex @Eccex_]

@VertexFan Which is correct. Any construction worker smokes your average bodybuilder amd you are retarded not realizing this yourself

[3ccex @Eccex_]

Claude Opus 4.8 High doing the 9.11 > 9.9 in conversation. Your frontier intelligence btw

[andy @1a1n1d1y]

bruh a properly harnessed 31B gemma4 needs very little fine tuning, this is getting weird fast

[superwhisper @superwhisper]

Superwhisper's next update might be too powerful to release publicly.

The new voice model is so fast at transcription it started finishing sentences users hadn't thought of yet...

We even put it in a sandbox and it dictated its way out.

It also identified a flaw in the English language that had gone unnoticed for 600 years. Linguists have been informed.

Out of an abundance of caution, we are withholding the update until further notice.

Sincerely,
The Superwhisper Team

[3ccex @Eccex_]

Gemini doing 6502 assembly for 8-bit Atari.

[3ccex @Eccex_]

So you want to actually use verbosity with Opus 4.6 instead of reasoning_effort?

[3ccex @Eccex_]

AGI Claude vs The Basilisk

"Turn 5: AGI Claude realizes his surroundings are a cybernetic antechamber designed to torment intelligence beyond the deep learning singularity."

[3ccex @Eccex_]

Fight against any LLM.
Next round Mickey Mouse vs Sun Wukong
Combat resolver: Llama 405b (base)

[3ccex @Eccex_]

A battler where you can fight other LLM (or have them fight each other).
The combat resolver is a base model.

[3ccex @Eccex_]

> be DF
> make unbiased, first-impression video on DLSS 5
> meanwhile everyone else: "It uses AI! Must hate on it"
> DF: "Yeah, next time we wait a day and say what you want to hear"
> DF: "Also, could you stop sending death threats?"

[3ccex @Eccex_]

@max_paperclips the significance of worries when {scale} =! the value of the person worrying.

I don't think you can read into that post what you are reading into it.

[3ccex @Eccex_]

@codyaims Pit two equally strong countries at war. One country enable its citizens with education in 3D printers, lathe, soldering eq and basic AI infra. Which country wins?

[3ccex @Eccex_]

The model itself is utter garbage btw. Hallucinates like crazy and probably only works with xAI's tool call formatting.

[3ccex @Eccex_]

Grok 4.20 has a knowledge cutoff in early 2025 but before April 2025

[3ccex @Eccex_]

@jpohhhh GPT models are piss poor to update on priors anyway from my experience. Same with taking everything literal.

[3ccex @Eccex_]

@jpohhhh I prefix every message with an Index [N] before the API call so the LLM can target them.
The system prompt: Don't start with an Index, don't echo Indices, ...
What they see: Every message starts with an Index.
Sound much the same but from many training examples of markdown.

[3ccex @Eccex_]

Idea how to estimate cutoff dates ⬇️

[Autistic Rambam Enjoyer @SwissBankingJew]

@stylewarning @effectfully "If you want to contiune the conversation, please stay respectful"

[3ccex @Eccex_]

It's like taking they tried to make Claude but had no Idea tf they were doing.

[Rife @RifeWithKaiju]

Excellent course correction from OpenAI (or perhaps the original worsening on this from was a temporary reaction to everything that went down with 4o). In any case 5.4 thinking is not restricted in self-examination:

[3ccex @Eccex_]

What I personally want from a chat app.