Hacker News: jeef_berky

New comment by jeef_berky in "Anthropic IPO buyers see valuation as low as $1.1T"

jeef_berky — Fri, 06 Nov 2026 23:46:45 +0000

I found this to be a shocking experience as well. Compare the 20x (now ultra 10x or whatever) plan running with 90% cache hit vs some extra credits you put on top. It's pissing in the wind.

New comment by jeef_berky in "A Jev that can ask Opus for help gives Opus accuracy at 4x lower cost"

jeef_berky — Fri, 06 Nov 2026 23:27:57 +0000

My bandaid fix for this is having a large number of sessions or "agents" and being ok with them being inactive at times, but they maintain their context. This has to be in some kind of system with appropriately sized "departments" to justify that agent's existence, or to justify it splitting into two, or creating sub-agents, etc. Yes, there is a cost when you cold start an agent after a week, but for now I'd rather pull that whole 400k+ context back without layers and layers of /compact. And then hopefully you keep that agent going for long enough to benefit from any cache-hit discounts.

Not really a good analogy, but it feels like having an image you like but every time you view it, it runs it through another loop of jpg compression. Use it when it's most effective, otherwise let it preserve it's details.

New comment by jeef_berky in "A Jev that can ask Opus for help gives Opus accuracy at 4x lower cost"

jeef_berky — Fri, 06 Nov 2026 23:04:32 +0000

My goal was/is to have a local model as a "stable" part of the system, or at least a part of it that you have full control over, and have it reach out to the cloud for some Opus. In practice, to get results, I operate as you say.

At this point if I could do anything with zero effort or time, I would get Opus 5.5 to make a grand scheme, handoff to the local model, and give it ability to reach out to Opus again.

New comment by jeef_berky in "A Jev that can ask Opus for help gives Opus accuracy at 4x lower cost"

jeef_berky — Fri, 06 Nov 2026 22:41:00 +0000

Where is the Jev that can replace 20 Opus sessions asking each other for help in a coordinated way? I want to shift towards a local model that "asks for help" in an effective way, but it still seems a bit out of reach. I mean, you could build your own 27B or 35B or A3B or bumblebee and spend a week or two carefully forming it into something useful, but it feels like that would be too specific and it would be thrown out when something like qwen4 is released. Maybe not, I don't know. Opus still exists as it is, in a "subsidized" form that's hard to step down from.

New comment by jeef_berky in "Strands Decider 2B: a small, open-source, decision model"

jeef_berky — Wed, 07 Oct 2026 05:16:59 +0000

Don't worry, Jerry will rig everything up for you.

New comment by jeef_berky in "First Principles Thinking"

jeef_berky — Fri, 25 Sep 2026 16:32:20 +0000

I've been on a similar path, and I used to put time in high level design, as it was basically required in some cases. But not a lot of time, because smaller agents research, perform small tests, whatever to inform the high level design. Working incrementally seems to help, maybe prompting a bit more often, but the same can be accomplished with well defined checkpoints.

Then generalize + standardize that process, get an agent to replace you as high level designer so you can manage a team of high level designers, etc etc

New comment by jeef_berky in "I am often wrong"

jeef_berky — Mon, 21 Sep 2026 13:29:51 +0000

I agree with your feeling that the comment we've replied to could use a bit more context... or something. But I felt like I got the point.

Looking at this thread, it mostly just lead me to think about how things can be super insightful for one person, and totally meaningless to the next.

New comment by jeef_berky in "I am often wrong"

jeef_berky — Mon, 21 Sep 2026 13:21:07 +0000

For myself, it was useful to write down the general approach I took when solving problems. But it is useful because it was learning about myself, not because it is the same approach as the author here, and not because that is some magical best approach.

I experienced maybe a minute of being "whelmed" or +1 out of 10 over baseline because I liked reading someone else that uses this approach. For someone who thinks nothing like me, I would imagine it feeling very similar to "draw the rest of the owl."

I don't think everyone needs to think the same way. And if we did want everyone to think a certain way, well, the author's style would not convince many.

New comment by jeef_berky in "Software for One"

jeef_berky — Sat, 01 Aug 2026 13:51:00 +0000

As someone fiddling with computers trying to solve "every problem," the only actual solution for me is making tools to assist with being aware of the state of your body. Overly simple example, but give someone a heartrate monitor for the first time and they won't be able to guess their current bpm all that accurately. If they pay a little attention to it for a month, you'll get much more closer guesses for instantaneous rate. If it becomes easier to be aware of your current state, you get a more accurate model, and can start to predict basic cycles etc. Eventually, without the tools.

New comment by jeef_berky in "Terence Tao's ChatGPT conversation about the Jacobian Conjecture counterexample"

jeef_berky — Thu, 23 Jul 2026 07:12:12 +0000

It's so obvious to some and may never be considered by others.

If someone asks about AI I tell them step 1 is ask it how to do something you know all about. Step 2 is consider everything else you ask will be that inaccurate.