2,117 karma · joined May 10, 2019
<a rel="me" href="https://hachyderm.io/@PeterBronez">Mastodon</a>
> The polar journey, like other zeppelin flights, was largely financed by stamp collectors; Graf Zeppelin carried approximately 50,000 letters sent by philatelists, and made a water-landing to exchange mail with the Soviet icebreaker Malygin, which itself carried a large quantity of mail sent by stamp collectors.
> Prioritize recall over precision.
Have you tried stemming your regex? That would help you catch messages where a different form of your word appeared. For example instead of “story” you look for “stor” which catches “stories” as well.
Then you might think, could we do an even better job by figuring out the general semantic intent of the query and history? Let’s project them into a semantic vector space! That’s an embedding.
Then you want to query that, which means you need a vector database. So now we can take the query, embed it, query the vector DB with that embedding and retrieve the N closest history documents. You can use that to augment the generation of the response to your prompt.
This is RAG.
Anyway, interesting to see different degrees of sophistication here. Certainly a handful of naive regex are very snappy.
There’s probably a hybrid approach where you use sophisticated NLP and embedding techniques to robustly define topics, then train a regex to approximate that well.
I’ve been using RapidMLX for this. The integrated speed tests matter because the quality of the backend is a moving target and the quantization / MLX format conversion also matter. It’s not enough to say “oh use this model family with X parameters” you have to add the architecture specific quantization too.
“Built on Python, Salt is an event-driven automation tool and framework to deploy, configure, and manage complex IT systems. Use Salt to automate common infrastructure administration tasks and ensure that all the components of your infrastructure are operating in a consistent desired state.”
https://docs.saltproject.io/en/latest/topics/about_salt_proj...
This is a way for web services to make your computer complete large amounts of compute at their behest. Tokens have value. There will be incentive for bad actors to use your local LLM for their own purposes, much like hostile crypto mining payloads.
This is an obvious target for prompt injection attacks and other malicious remote code execution. In many ways, model prompts ARE programs. The browser / local device would need to provide an LLM with the same sandbox guarantees as the rest of the browser. Can they be trusted to do that? Does anyone understand this well enough to do that with confidence?
I’m a big fan of local models, but I would be very cautious about letting random websites call the model I’m hosting on my local machine with open source software.
It’s not enough for a chat-first research agent, but it’s definitely enough to unlock features that rely on natural language understanding. Seems like a small thing compared to Claude/ChatGPT and the general hype, but still magic in its own context.
They have a metric called Model-Harness Index:
MHI = 0.50 × ToolCalling + 0.30 × HumanEval + 0.20 × MMLU (scale 0-100)
> Smart Cloud Routing > > Large-context requests auto-route to a cloud LLM (GPT-5, Claude, etc.) when local prefill would be slow. Routing based on new tokens after cache hit. --cloud-model openai/gpt-5 --cloud-threshold 20000
This electrify & integrate playbook has brought benefits to many industries, usually where better coordination unlocks efficiencies. Sometimes the smarts just add new failure modes and predatory vendor relationships. It’s showing up in space as more modular spacecraft, lower costs and more mission flexibility. But how is this playing out in manned space craft?
https://developer.apple.com/documentation/Updates/Foundation...
They released an official python SDK in March 2026:
Gets pretty meta when you’re evaluating a model which needs to evaluate the output of another agent… gotta pin things down to ground truth somewhere.
1) Government authored/sponsored materials are not considered “published” until they are available in both human and machine readable digital formats at no cost to the reader.
2) Publications of such materials
2.1) must not keep a record of individual usage beyond that technically necessary to provide access to the publication
2.2) not withstanding 2.1, take reasonable steps to secure digital publications from abuse that creates an undue administrative burden
In your example, the current crisis can be represented as:
A field exists and produces 1 unit.
A financial entity buys the field and applies unsustainable methods to increase production 100 units, keep 99.5 of them, distributes 0.5/n. People are pissed that they’re getting half of what they used to despite incredible productivity. The people elect a leader to fix the situation. The leader confronts the financial entity, and returns to the people with 4 units in their pocket and excuses.