Fable stuggles to attract users because it's so bloody risk averse it's ridiculous. At a hint of something that it might interpret as a red flag (e.g. cyber) it will downgrade straight to Opus. As someone who's utilizing LLMs namely in the context of infosec, Fable has simply been unusable.
Anthropic's best AI model struggles to attract users as cheaper tools thrive
571–580 of 740 posts
Re: Anthropic's best AI model struggles to attract users as cheaper tools thrive
#572People promoting investment in AI are fooling you with bad TAM estimates. For example, if we valued every spreadsheet created using the same metrics of pre-automation paper spreadsheets, spreadsheets would be a $100 trillion business.
Computing technologies are relentlessly deflationary. If the value of their TAM wasn't a fraction of a manual process they replace, they wouldn't have a productivity advantage. And the amount of TAM per unit often declines over the life of that product category.
I would be unsurprised to find investment in data centers to be 10X what was really needed. The same goes for where the LLM S curve starts to flatten.
Re: Anthropic's best AI model struggles to attract users as cheaper tools thrive
#573[dead]
Re: Anthropic's best AI model struggles to attract users as cheaper tools thrive
#574Earlier quoted context omitted.
And now I'm getting 10-20x as much done. I'd say the trade off is worth it. I'm struggling to scale myself even further. This tech is unreal and I have so many things I can do. For the first time, tech feels like the 90's-00's again. Everything is greenfield and exciting and big tech is struggling to figure out what to do about it. People are just hacking all kinds of stuff, and it's awesome. Feels like techno utopia…
I've seen this exact comment what feels like twice a day for the last 2 years, and not once have I seen the person making it back it up and show something even remotely impressive.
I dislike these comments just as much. What do you want people to show you? Most of us work on projects for other people where tasks that used to take a week take a day or less. It’s also a no true Scotsman as nothing we could show would be “good enough” because it’s just the same software engineering as before but faster.
But for instance I used to work in 1-2 client projects at a time and they take months now I can do 4-5 at once and they take a month. That’s a huge improvement
Re: Anthropic's best AI model struggles to attract users as cheaper tools thrive
#575This isn't the reason I'm considering leaving Anthropic. I don't think I can tolerate its writing style anymore. Reading Claude output is starting to cause actual psychological harm. I have tried many ways to get it to stop writing in its stupid punchy linked-in marketing-team voice, and I can't. Is there a model out there that sounds sound this awful? It's like rubbing sand into the folds of my brain.
I don’t think GPT is any better. I swear GPT-3 was the golden area of LLM prose. With a good fine tuning, you really couldn’t tell that an AI was writing (except for when it would devolve into utter nonsense lol)
Re: Anthropic's best AI model struggles to attract users as cheaper tools thrive
#576[dead]
Re: Anthropic's best AI model struggles to attract users as cheaper tools thrive
#577This isn't the reason I'm considering leaving Anthropic. I don't think I can tolerate its writing style anymore. Reading Claude output is starting to cause actual psychological harm. I have tried many ways to get it to stop writing in its stupid punchy linked-in marketing-team voice, and I can't. Is there a model out there that sounds sound this awful? It's like rubbing sand into the folds of my brain.
It’s really grating. It’s made technical analysis so tiring to read I often give up and give the output to another llm to make it readable.
Re: Anthropic's best AI model struggles to attract users as cheaper tools thrive
#578Where Anthropic f'ed up was treating their monetization the way they treat model training. Turns out that success in experimentation is not transferrable. They have tried to find the highest that the market pays for sota models; however, on the consumer side, this is just too confusing and unsettling: "You can only use Fable for a week as a part of your plan" "Be ready! You have to start paying per token!" "Nevermind…
Also, for whatever software engineering work I throw at Fable 5, Opus 5 also does fine. Apparently Fable is supposed to do better at long running tasks (in other words - burning more tokens without interacting with the user), but that's not the kind of work I'm doing. After Fable 5 launched, it was better than Opus 4.8 for sure. Then they rug-pulled Fable from me (EU), and later released Opus 5. Now I only reach for…
Fable was really excellent before the whole fiasco with the US government. Once they brought it back it was not the same at all. I have switched over to ChatGPT now for most things, its answers are way better than Fable in my experience. I still use Opus 5 for purely coding tasks.
This is why I think open weight models will win out in the end. Right now there’s too much going on behind the scenes with the models. Day to day you never know if you’re going to get smart Claude or dumb Claude.
Re: Anthropic's best AI model struggles to attract users as cheaper tools thrive
#579You know what. As a big Anthropic fan that pays €120 a month for Claude Max x5 and maybe €30 a month for API access I got extremely annoyed at the extremely variable quality of service I'm getting.
It is not only that every single turn with opus on high reasoning (high is the middle setting) takes at least 5-7min. It is barely usable interactively. Instead of a chat it feels like you're sending emails to it. Tasks that used to take an hour when it "reasoned" for 45s before it started doing anything now take almost entire day.
At least until few weeks ago it was horribly, mind bogglingly slow (a little better during US nighttime), but the quality was still good. I could not do things interactively, but providing prompts were fine it built stuff fine.
This is no longer the case. It makes stupid errors all the time. So you cannot leave it to complete some work, for example write infrastructure migration scripts a night before then you simply run the scripts and perform the migration during the day. Nope, every single script has stupid issues requiring use of the model to fix them. As they are written in it's own "spaghetti code" fixing them by hand is not an option.
It is clear to me they are doing some shenanigans behind the scenes to try and optimise their compute use. Either they quantized these models dynamically or do other things that affect quality.
In top of that they now do this stupid fingerprinting.
Anyone who knows how output vectors are turned into tokens knows it will eat up a lot of compute or destroy quality.
Re: Anthropic's best AI model struggles to attract users as cheaper tools thrive
#580nobody in my company wants to use opus 5 and fable is too expensive