ironically, I think this is why the jobs apocolypse is overblown, Ai is only good at a thing if the people using it are also good at that thing, and people are attributing Ai as superhuman at things they do not know themselves
I listened to a podcast on Ai and security today where they said they got access to a hackers workdir (after they were caught) and they had hacked 14 companies using GPT-5.2 and Claude <4.5 (forget minor). GLM-5.2 came up because, while not as good as Mythos, it's almost as good, i.e. you have to prompt more / cannot just give a fuzzy request and sit back. The harness likely matters more than the model
I also like doing that for topics that I am tangentially interested in. One minor thing that I find annoying is that the narrators switch roles in the middle of conversation. They start with the female voice explaining a concept to the male voice and suddenly they switch. In the meantime I have identified myself with the voice being explained to.
The nation is one terrorist attack away from rallying behind the president. And sadly the chances of that happening have gone up significantly in recent days.
How do you do taxes in Linux? Install a windows VM? I don't want to use the web version. With Google docs being good enough, I don't really need windows for anything else. Last time I checked, the tax software didn't run under wine.
I hope not because I’ve been doing my US taxes on Linux for 15 years.
It’s probably a specific windows desktop app, probably TurboTax by intuit, the company that lobbies to make filing your taxes hard and to destroy any free simple government app to file taxes.
So, not sure why they’d complain about not being able to help shoot their foot off but we all have preferences. :shrug:
Yes, it is the unfortunate reality of the US. You either fork over all your data to tax preparation companies on the web or pay them to use windows/MacOS software. There is still paper filing by doing taxes by hand, but that's a bit too inconvenient. Free online filing is restricted to $89k annual income.
Only if you think living longer is bad. Till early 20th century, urban life expectancy was lower than rural life expectancy in Britain. It was dubbed urban penalty.
Sometimes getting a second pair of eyes to look at the problem helps and is usually not a judgement of smartness of the first pair of eyes. Seems like it also applies to coding agents.
Indeed, I've also found that various models are good at various tasks, but I have yet been able to categorize "Model X is good at Y-class of bugs", so I end up using N models for a first pass "Find the root-cause of this issue", then once it's found, pass it along to same N models for them to attempt to solve it.
So far, which model can find/solve what is really scattered all over the place.
You are experiencing the jagged skills frontier. All models have these weird skill gaps and prompt phrasing sensitivity. This is the main problem solved by an llm-consortium. It's expensive running multiple models in parallel for the same prompt, but the time saved is worth it for gnarly problems. It fills in the gaps between models to tame the jagged frontier.
My very first use of the llm-consortium saw me feeding in it's own source code to look for bugs. It surfaced a serious bug which only one out of the three models had spotted. Lots of problems are NP-ish so parallel sampling works really well. Googles IMO gold and openais IOI gold both used parallel reasoning of some sort.
This is so true. Another thing, a model might be better at something in general, but worse if the context is too long. Looking at how GLM-4.5 is trained, on lots of short context, this may be the case for it.
GPT-5: Exceptional at abstract reasoning, planning and following the intention behind instructions. Concise and intentional. Not great at manipulating text or generating python code.
Gemini 2.5 Pro: Exceptional at manipulating text and python, not great at abstract reasoning. Verbose. Doesn't follow instructions well.
Another thing I've learned is that models work better when they work on code that they themselves generated. It's "in distribution" and more comprehensible to them.
The good old regression to the mean. Testing models as the second pair of eyes only when the first fails is going to give weird results... https://www.smbc-comics.com/comic/protocol
If you choose the worst performers, a part of their failure will be due to luck and part due to skill. Letting them retry will average the luck in that sample. (So likely increase) Of course you as the tester could be very unlucky and pull a sample of people with no skill who did get a bit lucky in the first place, but it's both unlikely and... it's a funny comic.
