bthornbury

Is the tradeoff of the parallel output that we don't get arbitrary string generation? like output # of tokens is fixed ahead of time?

Either way, really cool and impressive.

"Suddenly, waterfall is the thing to do."

Matches my experience. Like 90% of the work is the spec.

Except instead of weeks researching its like a few hours talking with an agent.

lots of good ones still

I discuss the testing approach and coverage with the model before and after, sometimes in a fresh thread that does a static analysis.

interestingly my input is still pretty important

codex pro plan currently

for me, almost all of the work is specs

I am no longer:

- reading docs for hours and hours

- typing (barely at all)

- writing code

- manually doing tight debug loops

- using an IDE

to do this I had to give up reading or even controlling the code and focusing on behavior/design-level control (not superficial, still dictating overall technical architecture)

i have agents doing everything from writing the code, verifying the code, hardening, increasing test coverage, analyzing behavior, algorithmic perf improvements, managing/deploying to cloud resources, etc... (pretty much everything)

and I am accomplishing projects that would take months or years in a fraction of the time.

that's way more than 10x.

somehow, this is harder and more cognitively demanding than writing code

we need some better standard long-context benchmarks.

needle in a haystack is not good for this, yes it proves the model can attend to its context, but in its usual form, somewhat trivializes the query-key relationship.

something like long-form Q&A would be more ideal. Like reading a book and answering questions that require synthesizing information derived from either the whole thing or disparate portions of it. Like describing an entire character arc in a 1000 page novel with examples and evidential moments.

the qwopus 27b model is good for grunt work style tasks, even across multiple files. Piping a bunch of things through, small factoring changes, stuff that just takes time to type out.

I wouldn't rely on it for large stuff like codex though. I haven't tried out deepseek/kimi, if we could run those locally it would be great.

promote yourself to PM only and use agents for authoring, verification, tests, checking the tests

orchestrator -> parallel subagents with investigation, authoring, verification, benchmarking subagents and integration / final verification handled by parent has improved my productivity too.

I feel like from here its agent swarms against a whole spec but haven't got there yet.

Still getting plenty of bugs in the more complex scenarios, but mostly (in some projects) i never have to look at the code and treat it like a black box

The classic papers describing diffusion are full of dense mathematical terms and equations.

For many (including myself) who haven’t stretched those particular math muscles since diff eq class a decade or so ago, the paper is just an opaque wall of literal Greek.

In this post I describe my personal understanding of diffusion models in less-dense terms, focusing on intuitive understanding and personal mental models I use to understand diffusion.

Something like a perplexity/log-likelihood measurement across a large enough number of prompts/tokens might get you the same in a statistical sense though. I expect those comparison percentages at the top are something like that.

AFAIK seed determinism can't really be relied upon between two machines, maybe not even between two different gpus.