jwpapi

Honestly the best red pill advice is:

1) Become world class experts in one thing 2) Become full on invested in LLMs learn on how to match general intelligence on business requirements (Evals, isolation, judging,etc..) I think this will probably be still 5 years 3) Get ready to do something in real life.

Reads like copium to me.

I personally dont like comments nor memory, nor runbooks. I use code and namings and constantly clean up.

I only do runbooks for things that I have to run again in the future and potentially mentioned (could be better in code, too tbh, but somehow I like to store these as runbooks, as sometimes infrastucture changes are mixed with a lot of explanation, scraping operations or something)

One off scripts are done in /tmp/

I have the lifestyle I want since I’m 25. I have nobody to tell me what to do and can work when I want on the things I want. I don’t really have to work actually. I just like to work

There was one guy commenting on my old post that stuck with me along the lines, what would you need for you to approve less. It first bothered me, but I kept it with me and I started thinking differently about my software and especially with Opus 5.5 + heavy linting and cleanup routines it seems sane to evolve to pipelinee engineering, but it also makes me feel that whatever I’m doing now is gonna be in the next model, because what I did yesterday is in the model now...

Reality is I just got turned. This is new for all of us and we all try to understand and learn this new reality. I always say what I believe most at that moment and I try a lot to not be a stuck with my old opinions, but rather have the best jugdement I can at time.

From experience my pendulum might overswing sometimes. Thank you for the reminder.

Are these highest stakes or most interesting. Highest stakes I would understand the ones that would have the most real-world impact.

If you self-host an open model on a sovereign hoster? They can only influence the responses correct? We are not talking about extracting information.

Or are we worried that open models send secret telemetry?

That is my point!

Yes that’s what I’m doing. Practically i have my competitors on lock. Most of the time security lacks and you can even figure out the model/prompts/database they use so the moat is gone.

However what you say is amongst devs, but that’s from my non-dev friend circle less than 1%. However but almost everyone has used chatgpt already.

I mean that is my point...

Sorry but Opus I can’t see the difference.

It’s definitely better than me in writing support emails!

I’m aware.

People seem calm in here, but I really now think is replacing a lot of industries. Like Website Designer. I totally see people asking chatgpt to build them a website instead of paying them 2,000 USD.

Soon it will connect to Ads and it will runs for it as well.

Honestly since Opus 5.5, Opus is my interface for 95% off my work. I don’t open emails anymore i let it summarize draft and handle most of the work and I only talk to Claude and not read/write the emails myself.

I also barely go in analytics dashboards anymore, Most I do is creating artifacts to explain if text is not enough for me to understand.

I’m going to bed. I have analytics running for every cancellation on my software to see what goes on. I have a/b tests running that are getting interpreted automatically. My bugs are getting fixed automatically.

I can monitor my competitors and copy every feature a/b test resolution they’ve done instantly for (3$ of my budget)

I’m an early adaptor of the latest models. The labs will have better tools already, but it really feels like a breaking point. What’s normal for me today is normal for more tomorrow and the friction has never been lower...

Interesting..

what advantage do you have with a gui multiplexer over the terminal?

Always the last model that gets announced is the best. The labs always know in advance.

I guess it depends on what you’re doing in my current work I couldn’t I have even with Opus 5.5 tons of iterations.

For example I was optimising a checkout experience and I at least did 30 iterations till I was happy.

But there is also other stuff like namings. They pick good names but if you work with exchangeable vendors I need even more explizit names

However of course I’m trying to get as much in linting and agents.md

But even if I would say Yes to everything I couldn’t come up with things to do fast enough.

But maybe skill issue ?!

This is one part of AI I hadn’t success with. I have very little need to run Agents over night, as my throughput is limited by my approval. Each work usually needs revisions, sometimes the bug is just a symptom of the root problem, sometimes I need to rethink how users want to use the app. Sometimes I need research.

I get how i prefer an already researched-version of a bug versus a raw bug notice, but I can do this with webhooks in the correct environment.

I really have no idea what to do with my agents over night. I can not build more. I can not think of more problems. My RAM is full.

A lot of people having different pricing experience. I think it’s important to understand that caching can differ, than if the agents spend waiting on code, or consume a lot of content. It depends on how you structure you codebase and how explorable it is, how much effort you set and probably some other issues.

For raw productivity most of what works is best and switching will cost you getting on use parity with other models, as you need to learn what they good at, potentially how the tool works and how to prompt it best.

For tasks that you implement in code, you should have benchmarks and evals.

That said for me was Luna a huge leap and 500+ of cost savings a month

I think going for more understanding is the way you need less understanding. The more solid your core understanding of your codebase is the less you need to know the details, the less missunderstandings the less iterations needed, the less mental capacity consumed

As this is an interesting story I think we need a change to HN algorithm.

The title is misleading, they don’t owe him.

The story is NVIDIA made a mistake and OP tried to exploit it and failed and is now grabbing another straw with that post.

It was not intentional shares to his or Nvidia’s understanding.

Everybody reading the actual story would not just upvote this.

The upvotes must be simple big corp hate +misleading title

I recommend you the book the rational optimist.

Not saying you’re wrong you just can’t be so certain

I don’t think its much cheaper like luna is double the cost. and you can easily use luna to generate a training set and train your own model. I think the main things its faster so a lot of people go the Wow moment I can classify my whole db at once. Or this model decides so quickly. Speed is the key differentiator, it’s a strong one, always.

I don’t want it to talk like me, I want it to talk exactly.

I don’t care to look up terms as long as they are correct.

All good I’ll check on my main!

And fine to do it in same folder same local laptop?

how do you get 2 cgpt pro?

Wasn’t there a website that was tracking that?

twice...