armcat

I asked ChatGPT and it answered Joe Biden. I used Google Search with Gemini summaries and got the same answer.

Regarding Google is it geo dependent? Like Lake Ontario?

Graphs are fantastic - PageRank [1] that powered Google Search (still does to some extent), TextRank for extractive text summarisation [2], Louvain for neighbourhood detection [3] all use graphs, as well as tonnes of others like belief propagation, Markov clustering, A*, and so on. So yeah, learn it.

[1] https://en.wikipedia.org/wiki/PageRank

[2] https://aclanthology.org/W04-3252.pdf

[3] https://en.wikipedia.org/wiki/Louvain_method

Back in 2018 I was working on a project for educating schools in certain parts of Africa and Middle East, on AI. The issue was it had to be air-gapped (no internet guarantees) and small/portable. I ended up downloading Wikipedia (not all of it), and then created a lightweight inverted index (for search) and TextRank (for summaries). The whole thing was on a Raspberry Pi, and installed in a small plastic enclosure. We had many of these and could easily ship them. There was an "AI Learning Module" as well we built, that started with Scratch, and then moved into Python. But if I was to do it all over today, I would probably add a tiny LLM and maybe a small ASR and a TTS model too, and you basically have a completely self-contained AI knowledge assistant in a cigarette box. The point being, you basically also have a "survival AI" you can take with you. Of course you still need electricity...

It's very unfortunate to see lack of any detail, model cards, or technical reports. We just know it's based off a Kimi.

A much better title for this would be something like "Cross-Tenant KV Cache Exposure in Shared LLM Inference: Analysis and Remediation". As others have pointed out there is no need to mention GLM-5.3 in the title. As someone that has been using GLM-5.3 I immediately thought there is a cause for concern here, only to find out there isn't.

They released the traces: https://huggingface.co/datasets/CohereLabs/10-2026-qwen-inve...

Completely agree, it's quite objectively and well written. I noticed Musk slipped into their X comments with a "concerned" emoji, and there is a chance that this will cause bit of hysteria and add fuel to the fire for "open models are bad for your health".

> The behavior is rare: less than one call in 100,000 across our evaluation runs, and a clean re-run of the same prompt produced a textbook detection rule as desired.

It's great that they caught this as it highlights the need for better sandboxing and guardrails. Not sure if this was bad intent from the model - the rarity of it feels more like some residue in the training data.

I used to. Apologies if it came on too strong, but it wasn't idealistic thinking. For print, I find that I can easily use AI to turn a web page into a nice clean LaTeX and offer that as a PDF for those that want to print or read offline.

I politely disagree. The original intent Tim Berners Lee had [1] was to create a navigable way of finding documents and information. Not printable. The entire issue we have today is that we have blurred both the navigable and the printable concepts. A website should just be a simple clean space for easily navigating and conveying intended information. To print, we have latex and other markup methods that are perfect for that.

[1] https://info.cern.ch/hypertext/WWW/TheProject.html

The same holds true for most classification tasks - if you have sufficient samples and high quality labels, training an xgboost or a deep learning model with a classification head is the way to go. You can use a frontier LLM on high-effort to help you with classification reasoning and explainability.

Sometimes the simplest ideas are the best.

One method I employed a lot "back in the day" for extractive summarisation of text, is TextRank [1], which applies that multi-billion dollar linear algebra procedure against text. I remember working at a news media company and testing TextRank against BERT for 2-3 sentence article summaries, and TextRank was better.

[1] https://aclanthology.org/W04-3252.pdf

You mentioned in your README that it was tested with GLM-5.3 but this is not a multi-modal model. Did you mean GLM-5.3-Flash?

The voice seems insanely immersive and uncanny. One issue that the full duplex models always have is tool calling. This is where GPT-Live [1] shines in the sense that you can use it as a fully practical agent. Looking forward to seeing how Dessa folks implement this.

[1] https://openai.com/index/introducing-gpt-live/

This is very cool! I will try your method for background and tile creation in the asset-generation tool I've built [1]. Would your procedural method work for animations as well, like sprites?

[1] https://github.com/acatovic/ai-game-studio

This is so awesome! How does it work with moving away from the mic, is it just random or does the conversational pattern lead it that way?

We replaced Looker with Hex [1], which was 7x cheaper (for our use case). But soon afterwards we built our own in-house data agent platform "Reef", which has now effectively changed how the entire company operates, from sales and customer success, to defining commercial and operational strategy. One of my engineers Samuel has done a small writeup about it [2]. Essentially, if you take care of your ETL and make data modelling a first class citizen, you can build your own BI (or even better - agentic BI) relatively easily.

[1] https://hex.tech/

[2] https://www.samuelsvensson.com/blog/reef

After reading the article I can see that the comments here are a bit off. For example see this quote from the article:

> It’s not dead. Some companies use it while I believe the majority of use cases are addressed by open-weight rivals.

The point the article is making is the incredible way Jev specifically caused insane hysteria and hype, and just as suddently faded away. Not that it's dead or that the concept is severely flawed.

Great writeup! I've done some work on Codecarbon [1] in the past, it's a pretty standard tool for measuring ML/AI workload emissions, it would be awesome to have your findings contributed to that project as well.

[1] https://github.com/mlco2/codecarbon

Thank you!

I’ve been doing this with voice mode in agent harnesses. In Codex for example I often speak orally and I find that (1) I discover something new myself, (2) agents seem to understand much better than pure text prompts. The only downside is it costs lot of tokens.

From my time building ML models (2016-2023) evaluation and interpretation was a large chunk of it. I often used "brute force" interpretation methods but there were others like SHAP [1] and LIME [2]. The other large chunk of ML work focused on feature engineering and data prep. Today in the LLM agentics era this is out the window. There are "evals" but no one does it seriously, and interpretation is often propagated to the consumer.

[1] https://shap.readthedocs.io/en/latest/

[2] https://christophm.github.io/interpretable-ml-book/lime.html

This is fantastic. I feel the KV cache system is basically a must for any serious inference hosting, both for the provider as well as the consumer. I guess they have this but just haven’t open sourced it yet.

Does anyone know if this is based on the NIST report on the "Evaluation of DeepSeek AI Models" [1]?

> DeepSeek models advance Chinese Communist Party (CCP) narratives much more frequently than U.S. models.

[1] https://www.nist.gov/system/files/documents/2025/09/30/CAISI...

I get "Sorry, this content is not available in your region." when trying to access this link. EU/Sweden.

Gibson's quote "the future is already here - it's just not evenly distributed" continues to hold true, especially today. Amazing author and unparalleled mind's eye.

I feel the hardware vibecoding is the next and currently untapped frontier. I’ve noticed AI labs are placing more emphasis on hardware as well.

Reinforces what Karpathy tweeted [1], that AI is effectively tackling 2-5% of work and the gap between those fully exploiting AI and barely touching it is growing. The question remains whether this AI focus on the small % of workers is actually intentional (by tech/labs), or a natural limit of where AI is right now.

[1] https://x.com/karpathy/status/2109361546505966046?s=20

The toaster is next!

"Reflection’s first model, Beam, scores 30 on the Artificial Analysis Intelligence Index, placing it #2 among US open weights models" [1]

[1] https://x.com/ArtificialAnlys/status/2111087606759030831