Viktor Malyi

Your team can deliver at least 2x more with AI and I will lead you there.

[Viktor Malyi @victormalyy]

what should I prompt Fable with 1% remaining?

[Viktor Malyi @victormalyy]

@sickdotdev Still massive roi

[Viktor Malyi @victormalyy]

idk what Slack whole team is working on right now but they seem to be missing the biggest opportunity of their lifetime.

correct me but it seems they have 0 useful AI capabilities in their product – "summarize a thread" is a joke and useless – I don't want the thread to be summarized, I want to take action that problem from the thread is solved while I'm away.

[Viktor Malyi @victormalyy]

@dexhorthy Yeah, just did this in 2 days

[Viktor Malyi @victormalyy]

My 10YO thinks it's Claude running on my laptop making internet at home slow.

No, it's just vodafone

[Viktor Malyi @victormalyy]

Frankly, I'm ready to pay 2x what I'm paying now for Anthropic's Max 20x plan if Fable is included with generous limits.

400USD is no brainer, because this is the only model you can delegate a substantial, big task and it will complete it e2e up to extent, where you need to come and finish the last 5% yourself.

I totally get this is not for everyone, but the enabler here would be to organize the process in a way that AI can understand the context and take action – and Fable is the currently most capable for that.

[Viktor Malyi @victormalyy]

fable 5 is insanely good at following long-horizon task plans precisely as they defined without any workarounds and shortcuts.

[Viktor Malyi @victormalyy]

funny to observe that Fable not only consumes its own weekly limit 2x quicker but also the general one as well exactly so fast.

[Viktor Malyi @victormalyy]

and you understand that the optimal solution lays somewhere in hybrid search but you need to experiment with hyperparameters (and you know how to direct claude on that).

And those skills is an exact differentiators between people playing with Claude and those who deeply understands the limitations and know how to workaround them.

Learn AI Engineering

[Viktor Malyi @victormalyy]

those of you who can AI engineering are the real superstars of today.

Let me explain why.

[Viktor Malyi @victormalyy]

because whatever model you use, the "AI Engineering" part of what's happening under the hood (or simply harness) is as important as the model itself.

And Claude Code/Cowork can't deal with 1.5-h call transcripts effectively right now: sometimes it loads the entire conversation into context, and sometimes use in-built tools like grep to just browse it.

Needless to say such a default way of processing large corpuses of text is not what you are looking for if you want reliable extraction of certain structured data.

And here come your AI engineering skills

[Viktor Malyi @victormalyy]

here we go Fable

[Viktor Malyi @victormalyy]

I just ran out of the entire 5h limit in just 1h with Fable 5 on 200USD plan only because I've asked it to "use search extensively" for a fairly easy task.

It went crazy with ~5 millions tokens spent and 101 Fable 5 subagents + and a crowd of adversarial subagents re-checking the results ran for ~1 hour and eating a good portion of my weekly Fable allowance.

So beware of doing deep research with Fable and not paying attention what you ask for.

[Viktor Malyi @victormalyy]

Been working with Fable 5 for the last ~8 hours and haven't been re-routed to Opus 4.8 even a single time.

Should I consider this as a success and a proof I'm using the model for correct use-cases?

[Viktor Malyi @victormalyy]

please someone fix Fable's verbosity.

The model DOES deliver the value comparing to Opus, but its' output is soo hard to read.

it feels like it tries to show itself as a more "smart" one instead of just presenting options or driving me in the right direction concisely.

[Viktor Malyi @victormalyy]

I wonder why the most very expert AI consultants push content targeted to newbies wanting to become AI consultants too instead of sharing their real cases, fails, tough challenges in AI?

[Viktor Malyi @victormalyy]

I see so much happiness from people after this announcement that Fable will be back again.

But what exactly is it going to change for you?

Yes, no doubt, that model is super-powerful, but have you really been constrained by Opus's performance?

Or is this rather about "I tried to one-shot this super-complex project (also asked it to make no mistakes) and Opus failed, so I'll give it to Fable and hope it'll pull it off"?