Hacker News Reader

Comment by leokennis | Hacker News Reader

From the perspective of “an average person”, ChatGPT is delivering fantastic products.

- For general chat and web search, occasional image editing, small coding work, document review etc. ChatGPT Plus is basically limitless and “just works” since 5.6. I’ve yet to give it some task it cannot do.

- When given sensible instructions, it hardly annoys with weird phrasing, glazing, or annoying constructs.

- The apps are very good (ignoring the initially terrible Codex app)

It’s easily my best spent $23 a month.

View on HN

You can get a lot of Codex usage out of that same sub on top of ChatGPT usage. Its a really good value and you can use that sub in any harness. In OMP I have Sol high as the orchestrator, Sol max as Planner & Reviewer, Luna max as task/coder. Very good setup. I'm on pro now and there are weekends when I use half a week's usage but I'll have 5 or 6 sessions going at once for many hours each day.

What is OMP?

oh my pi https://omp.sh/

This is quite interesting, I wasn't aware omp has a way to set up different models for planner/orchestrator/task.

/model then roles and also /agents for when a model chooses to delegate out sub agents

Thank you so much

I find that subagents usually burn more tokens and take longer, and produce about the same quality. A real killer use-case is using a VERY cheap subagent to do a lot of work, or reviews. Don't be fooled into thinking that a "scout" subagent will gather enough info for a "coder" agent to just start working.

OMP also does a lot of rewinds - where the orchestrator comes up with findings on its own and rolls back to a previous turn with a summary update. I don't know if it always chooses correctly between rewinds and sub-agents, but it seems to do a pretty good job. Luna is SO MUCH cheaper that even if it spends 30K overhead in context its still much less costly, and when they work in parallel its faster too.

Not even for the average person. Luna is an amazing model for most coding tasks.

> From the perspective of “an average person”, ChatGPT is delivering fantastic products.

It is a honeymoon still, enshittification is coming, who knows how that will look like given how much more expensive to run LLMs backed user experience. Some back of the envelope calculations: 300 million US users * 20$ a month * 12 months = 72 billion $ a year. 72B$ is some spare change for AI labs. That assuming entire US will be subs which is unlikely and outside of the US there are not many rich countries consuming it, India is the next market, then Brazil and Philippines I think, not super rich counties to say the least. I believe total revenue to just pay for the capex build out by the end 2027 should be on the scale of hundreds of billions a year.

Analysis totally excludes enterprise customer demand and or paid API usage which will only increase as apps integrate this into future knowledge work workflows.

Indeed, that is where the money is. Though enterprise is more focused on efficiently than retail and I'm not sure if they won't drift away from frontier models.

Depends on whether the frontier models can keep on offering better performance. The real efficiency is getting work done faster and better.

Compared to a $100k salary, a few hundred dollars a month is insignificant. If you can make the employee even just a few percent more efficient, it’s worth it.

> If you can make the employee even just a few percent more efficient, it’s worth it.

Are/were you in a position to make such decisions or it is a guess? I'm not but given certain evidence I doubt that few percent will cut it. I know some of the richest companies on the planet from SF Bay Area who won't give lunch for free to their engineers. So I'm not sure about "few percent" :-D 10x we were promised, now that is more interesting but we all know that 10x engineers is nonsense.

The cost of a full time knowledge worker in these companies is much, much, higher than the cost of a free lunch so there's more incentive here

How much efficiency is unlocked by providing free lunches? I was under the impression that was purely an expense.

On the other hand enterprises are gearing up to pay for shit like $99/user Agent 365, or paying for huge PTU reservations that go almost completely unused on weekends and holidays.

> ignoring the initially terrible Codex app

Still needs a LOT of work IMO.

I use a lot of ChatGPT remotez and 70% of the times is unusable and buggy (prompts disappear, a lot of errors, buttons don't work, etc.)

The current bugginess of Codex is the proof that OpenAI hasn't "achieved AGI internally" yet.

Why is that? Humans are considered AGI and we make godawful software 90% of the time.

Reply on news.ycombinator.com