I’m constantly disappointed by Anthropic’s models.
I become more disillusioned every day.
It seems like, by having the models write Python code, they tend to write Python code like an average developer. Which is to say, quite bad.
Add in the complete failure of the models to adhere to instructions in Claude.md, memory files, and added multiple times in prompts, I’m wasting huge amounts of time fixing bad design decisions that the model just slips in.
My own experience is that with the best models, if you provide reasonable tools, with progressive discovery, and a semi-clear user request, the system prompt is essentially worthless.
Yes, this is already accounted for in many benchmarks, but without deep context of the problem type, the top line pricing is the best starting point.
In my own experience, Fable is more token efficient than opus 4.8 with a higher likelihood of completing tasks correctly or at least with minimal corrective work. Opus regularly struggled to gather the correct context and reason effectively about what it had gathered.
GPT-5.6-sol crushes fable in speed and token efficiency and is clearly superior across many tasks that matter for me.
I also find all models from anthropic after opus 4.6 to suffer from the same ai slop language that long plagued OpenAI and seems to have been reduced drastically in 5.6
Non-competes in finance almost always come with compensation during the defined period.
The idea that a company can restrict at-will employee’s post-separation employment is absurd if they aren’t compensating the individual.
In many US states and countries outside the US, the enforcement of non-competes is very very hard. The problem is that they create a RISK of enforcement.
I'm a firm believer in this as well. You want to enforce a noncompete? Great. You have to pay me as much as what I was earning (plus benefits), possibly more, for as long as you want to enforce it. That amount should probably go up 50%+ per year too.
Executive level non-competes are probably the most damaging for the overall economy though.
If there's a market-dominating company, and execs are allowed to leave said company, start a competitor, get some investor dollars behind them, then start poaching employees from the old company, the market can have a really viable competitor quite quickly.
Without that ability, little monopolies spring up throughout the economy and use their size to crush upstarts, under-compensate their employees, overcharge their customers, and squeeze their suppliers.
Banning non-competes is an absolute requirement for free-market capitalism to function properly.
Salient arguments, although I am personally of the belief that limited non-competes make some sense, at least in the US, at least in some fields.
One of the other respondents mentioned one of the main issues with a DIY attitude towards modifying NCDs is the advent of digital signing of NCDs now, and I concur that NC documents really should be paper (though I can see an argument for adding a blockchain-like element/step with a digitized document that would capture the crossed-out sections). I used to deal with sections of NC forms I did not like this way myself.
A lot of people don't like non-competes but I think people give them a bad reputation sometimes. Not when they are being abused (like the case of the doctor and PE mentioned elsewhere, and as I said in some cases NCs make little sense to me; if it is isn't research-related, and/or doesn't involve some sort of patent or novel procedure or tech or research, it clearly makes less sense), but certainly when they serve to prevent people from running off and starting a new company and competing with the company they just left (e.g., how the AI field especially is getting very glutty and 'competitive' now; it resembles less of a free-market and more like Battle Royale in ways).
Pretty sure non-competes prevent economic collapses, layoffs and bankruptcies for many companies.
The danger is abuse of them, not their existence. I don't think banning them outright is good. Especially in Fintech, which is a field rife with moral and ethical quandaries.
I do want to point out that it irks me when people make a big deal complaining about a non-compete afterwards, when they know they signed a non-compete prior to a meeting or job or role somewhere. Part of the point of non-competes is that generally the people that want you to sign them know there is a reason you would want to talk about something or use it elsewhere.
Abuse should be prevented and preventable by both sides of the non-compete and NDA processes though.
Should someone be prevented from working with models after working with them at another company? Maybe. If you sign and say yes and you are paid (even if it is not after you leave; you were given notice, though clearly you may not like it; the ability to renegotiate should exist before leaving), then, like it or not, you signed it. But a better approach is just more specific NC clauses (specific sorts of models or specific fields being ruled out, not the whole shebang).
But as I said, I see good reasons for them existing.
I gave fable 5 a task for which opus has been really really underperforming. Fable 5 took far less time and produced actually useful analysis. Instead of just regurgitating roughly what the code already does or misunderstanding entirely, it identified multiple routes to improve. Now, the code it is analyzing is not very good as it was mostly produced by opus.
Opus had consistently ignored my instructions and looped on broken logic over the last several weeks.
I’ll be sad when this model is removed from Claude code because I won’t be paying api pricing to work on open source projects.
As with anything, it depends on the quality and skills of the people doing the designing. I have yet to really work with a designer that embedded themselves enough in the UX to design anything I thought was good.
Granted, I have not worked with that many designers so it’s maybe small N problem.
The gap between figma and where our front end teams work and design is still too large.
I think the more likely result of this is fewer designers and more work for developers with some design skills.
I become more disillusioned every day.
It seems like, by having the models write Python code, they tend to write Python code like an average developer. Which is to say, quite bad.
Add in the complete failure of the models to adhere to instructions in Claude.md, memory files, and added multiple times in prompts, I’m wasting huge amounts of time fixing bad design decisions that the model just slips in.