Viewing profile — popinman322

With RAM prices how they are right now the used laptops might be more valuable than $1500.

Very much agree. Until the vibe-coded version has been fully audited and profiled to perform, within reasonable tolerances, as well as the original code base, it feels like a bad i…

Oh, this is great! I've filed bugs with JetBrains before and had them take months getting to my ticket, often with multiple hand-offs between team members; being able to provide a …

Does anyone know whether we'll be receiving transcoders for this batch of models? We got them for Gemma 3, but maybe that was a one-off.

I've found that Gemini models often produce pseudocode that seems good at first glance but is typically wrong or incomplete, especially for larger or more complex functions. It mig…

It doesn't look like the code anonymizes usernames when sending the thread for grading. This likely induces bias in the grades based on past/current prevailing opinions of certain …

They're comparing against open weights models that are roughly a month away from the frontier. Likely there's an implicit open-weights political stance here. There are also plenty …

You can swap experts in and out of VRAM, it just increases inference time substantially. Depending on the routing function you can figure out all the active experts ahead of the fo…

The executive branch is currently ignoring the law. Why would they start following it in 2029?

Not a fan of censorship here, but Chinese models are (subjectively) less propagandized than US models. If you ask US models about China, for instance, they'll tend towards the anta…

Assuming you're doing local inference, have you tried setting a token filter on the model?

DeepSeek was built on the foundations of public research, a major part of which is the Llama family of models. Prior to Llama open weights LLMs were considerably less performant; w…

It's always very interesting to see people pull out threads with low like counts (like 12k) and claim that central idea of the post is widely held. We're talking about platforms wi…

The LSP is limited in scope and doesn't provide access to things like the AST (which can vary by language). If you want to navigate by symbols, that can be done. If you want to kno…

Google Trends make it seem like we're out of the exponential growth phase for LLMs-- search interest is possibly plateauing. A decline in search interest outside of academia makes …

Try LiteLLM; their core LLM proxy is open source. As an added bonus it also supports other major providers.

Huge +1. If I'd understood this mantra earlier in my career it would have saved me a large amount of hassle. For juniors: any time you send something important to your manager, con…

Grep is also useful when IDE indexing isn't feasible for the entire project. At past employers I worked in monorepos where the sheer size of the index caused multiple seconds of de…

It's possible to get stuck in merge hell where all your reviewers ok the PR but someone merged a conflict 2 seconds ago, or you've got a reviewer in Singapore while you're in SF an…

This is where supporting machinery & RAG are very useful. You can auto- lint and test code before you set eyes on it, then re-run the prompt with either more context or an altered …

Tangent here: really? I've found base Whisper has concerning error rates for non-US English accents; I imagine the same is true for other languages with a large regional mode to th…

Also, similar to Orca-Math but without a teacher model. They also followed an iterative DPO/KTO scheme, but with no length normalized NLL loss term.

I remember stumbling upon an early discussion about this [0] a bit ago in the EleutherAI discord when searching for discussion about a paper; I'm glad to see it's turned into somet…

Any news on when Unsloth's parallel full tuning will be available?

You could likely also combine the LLM with a coverage tool to provide additional guidance when regenerating the fuzzer: "Your fuzzer missed lines XX-YY in the code. Explain why you…