Posts by matjaggard 395 publicly visible posts • joined 22 Apr 2015
When discussing the price, why wasn't this compared against cloud offerings of open weights models? In my experience running a model in one of the major or local cloud providers allows you to specify your data stays in your own country for a fraction of the cost, even spread over lots of years - and who knows how long this hardware will last / be useful for.
@cyberdemon I was with you until you said that it's useless for defenders. It is absolutely not - it can find an fix almost all well known classes of bugs and that is enough for most of the world. Very few security incidents are caused by something novel. It's also got very good recently at not breaking other code now by using semantic search and abstract source trees.
I totally agree with the "come off your high horse" comment. "Reasoning" is a perfectly valid technical term for when an LLM talks itself through a prompt before responding back to the user. Is this human reasoning or human intelligence - no, and nobody is claiming it is - but re-using human terms for what a machine does is both perfectly reasonable and as old as computers themselves. They can't "calculate" like a human, or "analyse" like a human or "decide" like a human but of course with have decision trees in programs, analysis of data, even in simple if/else code, calculations.
That's not a great reason to avoid crypto. "Real money" doesn't have any basis in real value! Best to avoid it due to the types of people who are pushing it, the corruption, the reliance on ever increasing amounts of compute power (in some cases), the lack of backing and guarantees by your nation state, but not due to the basis which is missing from all currency.
You're quite right that this is not a good way of getting quality data - mostly because of the increase in Claude users during that time. Also as time goes on the people using Claude will likely be less technical (at least as far as my company goes anyway).
Of course the prices are going to increase, that's a given, but I think you're underestimating the users of Claude. I think the majority will know what a token is and whilst they probably don't, they could estimate how many of them they're using. Anthropic are playing a slightly dangerous game by making their costs so opaque - people are not stupid, they will build things to replace Code, Co-work, etc. if the pricing goes too high.
Well aren't you all pleasant people!
A lot of people using these are blind - they have small speakers in the arms and so a bind person can suddenly be able to read signs, know what's around them, call a family member if they feel confused or concerned.
But instead of reading a story like this and thinking "these tech companies should stick to the law" you just call for them to be illegal or worse, "euthanise" the users.
You should be ashamed of yourselves.
Am I surprised the tech companies need a wrap on the knuckles? not at all but that doesn't mean we should ban innovation.
I think you're wrong, I think despite us not paying them like they'd get in the private sector lawmakers are at least 90% *trying* to do the right thing but sadly the odds are stacked against that outcome. They don't have the tools they need to make the right laws so they have to take advice and that advice generally comes from people with enough money to get to lobby them.
The whole funding model, choice of politicians and lobbying rules all need to change to fix this. Thankfully here in the UK we do the bare minimum of limiting political spending but it needs to go so much further.
This is not vibe coding. This is an expert using a new tool like Clause or Copilot.
Having a AI write code for someone who fully understands and problem domain but can't code, and is willing to sit and chat to a tool for hours can produce something pretty good. That's Vibe Coding. Outputs can be functional, you can start a business with it before needing to re-architect at some point. The code is actually pretty well laid out and well written if you can ignore the excess of comments and emojis.
Our current issue is how to deal with this in the longer term. Do you have a moment when it's thrown over to proper devs? We've learned from old methods of testing and hours of Jira ticket writing that throwing things over the wall to the next team doesn't work well but how do we handle this if not?
My company uses it and we all hate it. It's very fast to get something working and very slow to keep it working because of all the magic it does. Things just magically get changed from camel case to snake case for example, types are impossible to follow and can change as you use them due to the "becomes" keyword, and Robocop is more Nazi than police by default.
I think you might have a touch of confirmation bias in your reading.
I also dislike black boxes in my code but I don't mind LLMs because I know how they work and from experience where they don't. Probabilistic output is not quite the same as a black box.
I've not found it any harder going back to code I've accepted from an LLM vs code I wrote myself. I just don't accept anything bad from the LLM.
AI doesn't solve all problems, probably not more than 5-10% but that doesn't make it useless. It makes it a tool that you can learn to use and those who do will likely succeed more than those who don't in the long term.
Your logical process there has a mistake. Someone competent enough to spot a mistake might well want to use AI for a number of reasons. For example LLMs are incredibly good at finding salient points in masses of data - documents, code, etc. Given the correct task it can save huge amounts of time and for many problems, checking it is trivial.
Why are people focused only on individuals who should be dealt with whilst Fujitsu are STILL winning government contracts??! I'm not saying individuals don't need to face justice (but not violence - come on people) but we can't ignore the continuing decisions to pay more and more to Fujitsu.
I think the use of the phrase "nobody asked for" in saying what Microsoft are doing is to miss how business works. Tech businesses that *just* do what people ask for will fail because anyone can listen to people and produce something that fits their stated needs. The difficulty is fitting their actual needs or doing something brand new that people haven't thought of.
That's where AI is currently and it will change our entire world - especially for developers who need to understand and eventually embrace it or may as well quit.
I doubt it's that explicit or sinister, Hanlon's Razor is worth remembering.
The issue here is that none* of our MPs have the required technical understanding to properly scrutinise policy. I saw once a finance guy saying that everyone either has power or understanding of how to solve issues, never both. I think it's our popularity contest version of democracy plus the fact that we don't pay MPs enough to attract clever people (unless they're clever enough to know they can help a mate whilst an MP and get a lucrative job afterwards) that causes the issue.
*a guess, not fact checked.
I'd agree with most of that, security is hard.
The first sentence I'm not convinced by. Firstly because using a system in a way not intended is frequently not a security issue and is often very valuable. Secondly it's sometimes the requirements that cause the security issue.
I don't think AliExpress itself is an issue, the stores on there vary hugely but almost all of this type of hardware is made in China anyway.
The router you linked to is more expensive, has all ethernet ports at the slower speed (so to get almost equivalent you could buy a cheap gigabit hub) and as far as I can tell it's not FCC compliant, but it does have other great features like the 3 mPCIe interfaces. I think it's really designed for a slightly different use case.
