I’m using Opus 4.8 with ‘xhigh’ effort level. I’d rather the little extra time (and tokens) it takes to get the best answers, esp as regards to development/coding. I’m on the Pro plan, which costs about the same as a cup of coffee per week. So far, I’ve not really come close to exceeding the usage limits.
It’s a shame the newest Fable 5 model has been disabled by the US government, on the pretext of security vulnerabilities. I suspect it’s more a backlash due to Anthropic not allowing it’s models to be used for military purposes, but I don’t want to get into politics.
For my DSA purposes, Claude drew upon both the polyorb and garlic codebases. It also examined (and referred to) this forum . It was amusing to see Maxim, Darek and even myself mentioned by Claude.
Absolutely ! I suspect that, given a set of rules, it could nicely format an entire code base in minutes. Another probable use case is as a bindings generator. To think of the many months I spent (wasted) developing the (largely unused) swig4ada binding generator.
Frankly, I fail to see any ethical problems, other than job losses, but in that case you may as well say the industrial revolution was unethical.
Yes, and that was the main reason I initially resented AI. “My favorite hobby is being destroyed. I am becoming obsolete.”. Now, I’m seeing it more as a shift in focus. The game is now becoming precision of requirements specification by human (the 1st phase of the s/w lifecycle) and automating thru AI the remaining phases (analysis, design, implementation/coding, testing).
I guess we are not quite there yet but the advances in only the past few years are staggering.
Some new models have been released since and the ones that were already present have improved quite a bit. I am very impressed with DeepSeek-v4-flash. It is brutally cheap, fast and it can do basic contract reasoning now! GLM-5.2 is a beast, but it comes with a substantial cost, it will probably improve in the next few weeks.
Also, the tool that I use OpenCode, made their Data public (only the “Go” plan, so the flat, 10 €/month system), which you can find here https://opencode.ai/data/ It shows the cost per token, what is being used more, etc. It does not have the State-of-the-art models, as those are not included in the “Go” plan. Most of the models are open, however, which is something that I like. I think some of you would like it.
I’m delighted to see you’re making progress with your DSA endeavour! And I can relate to your amazement; been there myself.
Maybe you can make use of my ZeroMQ implementation. So far, it has been propelled only by my own Podmander project. I’d love if it got more adoption and useful feedback to improve it.
. It was amusing to see Maxim, Darek and even myself mentioned by Claude.