Miyamoto Musashi

developer/builder. My twitter alter-ego for AI/LLM related stuff.

[Miyamoto Musashi @solminded]

@DavidSacks @martin_casado trusted partner came up with a jailbreak huh? which partner is that? did they came up with a jailbreak before?

[Miyamoto Musashi @solminded]

@AnthropicAI Well you were probably going to nerf it in a week or so due to high demand. But I must admit it was quite good experience while I was working with it for the last few days.

[Miyamoto Musashi @solminded]

@elder_plinius @AnthropicAI thx obama

[Miyamoto Musashi @solminded]

@deedydas Did you try the same tasks with GPT 5.5 or any other model using a proper loop + context? None of these tasks sound impressive in 2026 ifmyou have a good harness and workflow tbh.

[Miyamoto Musashi @solminded]

Hey @OpenAI all codex cli and codex app returns the same error response (with 5.5 model only)

{ "type": "error", "error": { "type": "image_generation_user_error", "code": "invalid_value", "message": "The model 'gpt-image-2' does not exist.", "param": "tools" }, "status": 400 }

[Miyamoto Musashi @solminded]

@sama I have a better term. gamblemaxxing.

[Miyamoto Musashi @solminded]

@EHuanglu and what is the ROI on that 50$ spent?

[Miyamoto Musashi @solminded]

@Saboo_Shubham_ @Teknium “space time warp visualizer” is the goal?

[Miyamoto Musashi @solminded]

@zan2434 @ashpreetbedi @eddiejiao_obj @drewocarr the query was “cats”. failed to build full version and draft version was pretty.. drafty. good concept but it needs a lot of work to be usable

[Miyamoto Musashi @solminded]

@deedydas it was gemini obliterating the SWE benchmarks last time yet the real world usage was not good. benchmarks does not mean much these days.

[Miyamoto Musashi @solminded]

@LLMJunky @bensig Even if MemPalace itself stays local, once you connect it to a commercial model through MCP, retrieved snippets can be sent to that provider as part of prompts or tool-augmented context.

[Miyamoto Musashi @solminded]

@jordihays @0xluffy why though? control the narrative?

[Wes Bos @wesbos]

Claude Code leaked their source map, effectively giving you a look into the codebase.

I immediately went for the one thing that mattered: spinner verbs

There are 187

[Miyamoto Musashi @solminded]

@adisingh https://t.co/QOc5pGWn1s

[dax @thdxr]

please shut the fuck up i don't even care about the specific thing you're saying i'm just so tired of hearing predictions one after the other telling me what the future is going to be like just please shut the fuck up

[Miyamoto Musashi @solminded]

@soraofficialapp well that was expected I guess. high costs, better competition, no real moat.

[Miyamoto Musashi @solminded]

crewai, hermes, openclaw.. if you are using any of these or an agent framework make sure not to have installed version 1.82.8 of litellm

[Andrej Karpathy @karpathy]

Software horror: litellm PyPI supply chain attack.

Simple `pip install litellm` was enough to exfiltrate SSH keys, AWS/GCP/Azure creds, Kubernetes configs, git credentials, env vars (all your API keys), shell history, crypto wallets, SSL private keys, CI/CD secrets, database passwords.

LiteLLM itself has 97 million downloads per month which is already terrible, but much worse, the contagion spreads to any project that depends on litellm. For example, if you did `pip install dspy` (which depended on litellm>=1.64.0), you'd also be pwnd. Same for any other large project that depended on litellm.

Afaict the poisoned version was up for only less than ~1 hour. The attack had a bug which led to its discovery - Callum McMahon was using an MCP plugin inside Cursor that pulled in litellm as a transitive dependency. When litellm 1.82.8 installed, their machine ran out of RAM and crashed. So if the attacker didn't vibe code this attack it could have been undetected for many days or weeks.

Supply chain attacks like this are basically the scariest thing imaginable in modern software. Every time you install any depedency you could be pulling in a poisoned package anywhere deep inside its entire depedency tree. This is especially risky with large projects that might have lots and lots of dependencies. The credentials that do get stolen in each attack can then be used to take over more accounts and compromise more packages.

Classical software engineering would have you believe that dependencies are good (we're building pyramids from bricks), but imo this has to be re-evaluated, and it's why I've been so growingly averse to them, preferring to use LLMs to "yoink" functionality when it's simple enough and possible.

[Miyamoto Musashi @solminded]

@mattwelter more slop more slop non stop

[Miyamoto Musashi @solminded]

@ThePrimeagen can you call it a theft if it was already stolen

[Miyamoto Musashi @solminded]

@burkov they make money over engagement so they will be prompting all day to find attention bait post ideas

[BURKOV @burkov]

LLMs process text from left to right — each token can only look back at what came before it, never forward. This means that when you write a long prompt with context at the beginning and a question at the end, the model answers the question having "seen" the context, but the context tokens were generated without any awareness of what question was coming. This asymmetry is a basic structural property of how these models work.

The paper asks what happens if you just send the prompt twice in a row, so that every part of the input gets a second pass where it can attend to every other part. The answer is that accuracy goes up across seven different benchmarks and seven different models (from the Gemini, ChatGPT, Claude, and DeepSeek series of LLMs), with no increase in the length of the model's output and no meaningful increase in response time — because processing the input is done in parallel by the hardware anyway.

There are no new losses to compute, no finetuning, no clever prompt engineering beyond the repetition itself.

The gap between this technique and doing nothing is sometimes small, sometimes large (one model went from 21% to 97% on a task involving finding a name in a list). If you are thinking about how to get better results from these models without paying for longer outputs or slower responses, that's a fairly concrete and low-effort finding.

Read with AI tutor: https://t.co/MipHHO6rjX

Get the PDF: https://t.co/XQrqiaGwIO