Where the AI market is talking

Xiaomi: New CPU matches Apple cores single threaded, much faster multithreaded

I think describing what's happening "other countries seem to sit on their hands and happily outsource every bit of manufacturing to a single country" is disingenuous. It's been dubbed China Shock 2.0 and it's being taken very seriously by the EU. The US admin is a joke and too busy shooting itself in both feet to have coherent policy https://en.wikipedia.org/wiki/China_Shock_2.0

How Europe is killing makers and micro-entrepreneurs

> eu expanded too much and grew too large to actually have a coherent purpose. Anybody seriously thinking that seems completely oblivious to how central EU is to the economies and every day lives of member states. There is a lot to criticise about EU but without it, the Europeans lives would be much worse and their economies would be nothing to speak of. It is very sad that populists managed to popularize the image of the bloated unnecessary monster. Even otherwise intellectually capable folks.

I were 17, I'd learn how to build LLMs from scratch

What about doing abliteration, weight pruning, representation engineering, etc, directly to open LLMs instead? Building an LLM from scratch has a hard split between a tutorial project you can complete in a weekend (that's useless for actual usage) and then a solid 1km high brick wall if you want to create anything actually useful from scratch. Modified open models have a very active community around them, without the need to look much further than Hugging Face.

I gave Qwen 3.8 27B a reverse-engineering job and it finished in 30 minutes

> As it turns out, probably unsurprisingly, Qwen recognizes common jailbreak attempts, and one of the first things it told me was that it wasn't going to fall for the jailbreak prompt Now also see latest submission, https://news.ycombinator.com/item?id=49409073 : # I spent $266 and four AI models to own my tablet. GLM-5.3 finished it in a day > Quick context: the tablet is a 2021 Fire HD 10 that ran my Home Assistant dashboard and kept powering itself off: the logs showed Amazon's own software issuing the shutdowns, and the only permanent fix was root, which has never existed publicly for this model. Anthropic's and OpenAI's cyber safeguards wouldn't touch the project Why should Anthropic and OpenAI thrive: they do not work on real problems.

I gave Qwen 3.8 27B a reverse-engineering job and it finished in 30 minutes

> As it turns out, probably unsurprisingly, Qwen recognizes common jailbreak attempts, and one of the first things it told me was that it wasn't going to fall for the jailbreak prompt Now also see latest submission, https://news.ycombinator.com/item?id=49409073 : # I spent $266 and four AI models to own my tablet. GLM-5.3 finished it in a day > Quick context: the tablet is a 2021 Fire HD 10 that ran my Home Assistant dashboard and kept powering itself off: the logs showed Amazon's own software issuing the shutdowns, and the only permanent fix was root, which has never existed publicly for this model. Anthropic's and OpenAI's cyber safeguards wouldn't touch the project Why should Anthropic and OpenAI thrive: they do not work on real problems.

I gave Qwen 3.8 27B a reverse-engineering job and it finished in 30 minutes

I still think that Anthropic went the wrong way. It would have been much more entertaining to ask the model to find a non trivial zero not on the line and give it encouragement. To see what exactly it will come up with.

OpenAI leader warns of threat of 'persistent' AI cyber-attacks

I gave Qwen 3.8 27B a reverse-engineering job and it finished in 30 minutes

Yeah, the Chinese government used the same method last year to hack the US government using Claude Code. Making each piece of work small enough to be plausible. Compartmentalization. (Also saying "nah it's cool I have permission", heh) https://www.anthropic.com/news/disrupting-AI-espionage

Why aren't my two Cortex-A9 cores cache coherent?

If the hard cores are configured with incorrect cache coherency options you may not be able to correct the issue in fpga fabric. But this is pure speculation, I haven't checked Cyclone V documentation myself.

I gave Qwen 3.8 27B a reverse-engineering job and it finished in 30 minutes

Hugging face is filled with uncensored versions of your favorite local models, so in a way they are shipped without the refusal stuff, via the magic of fine tuning or however they get this stuff out of models.

AI Labels Are Big Tech's Most Basic Responsibility, Even Those Claude Watermarks

who benefits from a watermark something like anthropic can only detect?

Anthropic's best AI model struggles to attract users as cheaper tools thrive

Opus 5 is not consistent. When it is better than 4.8, it's much better. But when it's worse than 4.8, it's much worse. I am not sure what they are doing, but I prefer the more consistent performance of Opus 4.8. I am seriously considering dropping my Anthropic subscription and upgrading my OpenAI subscription. Fable guardrails trigger on the most innocuous references. I am working on a model design, I based the design on the cortex. If I use biological terms at all, no fable for me. I have had to remove documentation and knowledge docs from my repo to scrub all biological terms, and still it will sometimes make the connection on it's own in the middle of a session and drop me cold. I don't understand how they could burn through all the goodwill so fast, but I have gone from a fan to seeki…

Anthropic's best AI model struggles to attract users as cheaper tools thrive

Opus 5 is not consistent. When it is better than 4.8, it's much better. But when it's worse than 4.8, it's much worse. I am not sure what they are doing, but I prefer the more consistent performance of Opus 4.8. I am seriously considering dropping my Anthropic subscription and upgrading my OpenAI subscription. Fable guardrails trigger on the most innocuous references. I am working on a model design, I based the design on the cortex. If I use biological terms at all, no fable for me. I have had to remove documentation and knowledge docs from my repo to scrub all biological terms, and still it will sometimes make the connection on it's own in the middle of a session and drop me cold. I don't understand how they could burn through all the goodwill so fast, but I have gone from a fan to seeki…

Fast and Hard Code

Claude Code more or less does have the tools to do this: plan mode, todo lists, user question prompts, et al. What it does not have is a "guided" mode where the agent (or harness) interviews you and helps you structure a work plan for the agent, including eliciting those success criteria and any design constraints the user might have in mind (eg it will be used on a boat over slow satellite connection). I can't speak for OpenAI but I get the impression that Anthropic think of these things as opt-in power user features, perhaps on the premise that their LLMs alone are "smart enough".

Fast and Hard Code

Claude Code more or less does have the tools to do this: plan mode, todo lists, user question prompts, et al. What it does not have is a "guided" mode where the agent (or harness) interviews you and helps you structure a work plan for the agent, including eliciting those success criteria and any design constraints the user might have in mind (eg it will be used on a boat over slow satellite connection). I can't speak for OpenAI but I get the impression that Anthropic think of these things as opt-in power user features, perhaps on the premise that their LLMs alone are "smart enough".

I gave Qwen 3.8 27B a reverse-engineering job and it finished in 30 minutes

There are versions of Qwen3.8-27B that are unrestricted and available from hugging face. "It will comply with harmful, unethical, offensive, or illegal requests that the original Qwen3.8-27B would refuse. It has no meaningful built-in guardrails."

I gave Qwen 3.8 27B a reverse-engineering job and it finished in 30 minutes

Local models would be even better if they did not ship with all the refusal shenanigans built-in. You can safely bet organized crime has access to the best models without these hoops, which makes the case that the average user (=non-criminal) should have access too. As I understood from an ex-Anthropic employee, some orgs got access to Mythos based on their high enough spending level, not on other grounds. Either we are in command over the software, or the corp is in command over us via the software. I can on a theoretical level understand the concerns, but either we ban all LLMs or we have a level playing field for everybody. Let's not forget: defense and offense are different sides of the same coin in software. I guess this wouldn't apply to bio weapons, but I am not in the know about t…

I gave Qwen 3.8 27B a reverse-engineering job and it finished in 30 minutes

My M5 Pro gets around 12-15 (6 bit MTP), although I haven’t worked on optimising it at all yet. A nice thing about running locally is you can run an uncensored model and you don’t have to worry about TOS violations on your OpenAI account when you ask it to “reverse engineer this ancient router firmware and give me a licence key that will work on it”.

I'm becoming AI-blind

Again, you keep mentioning intelligence but you talk more like you're talking about awareness/'understanding'/semantics. I agree that these are tricky things to define precisely but there is a reason they are different words. It's also odd considering you mentioned 'blindsight' which is a whole book about the idea that you could have intelligence but not awareness (something I am not sure is a coherent concept, personally). And yeah, part of it does seem to stem from disagreement about definitions. To me Searle seems to assume much more in his definitions as obvious than he explicitly states, which is why the Chinese room seems like such a non-argument from my point of view.

Why your local LLM feels dumber than it is

I believe whoever lays down the best structure around LLM will take the lead. Proven in Anthropic vs OpenAI.