rgbrenner

6 ms·

31.
by rgbrenner 1y ago
openrouter requires an openai api key.
32.
by rgbrenner 1y ago
the media but also the llm providers actively encourage this to fuel their meteoric valuations that are based on the eminent value that would be provided by AGI replacing human labor. the entire thing — from the phrasing of errors as “hallu
33.
by rgbrenner 1y ago
why would the llm share any of the blame? it has no agency. it doesn’t “understand” anything about the meaning of the symbols it produces. if you go put your car in drive and let it roll down the street.. the car has 0% of the blame for wha
34.
by rgbrenner 1y ago
pro is the $20/mo plan that they recently started allowing access to claude code.. but i’ve heard users hit the rate limit with a few queries.. so imo that sounds about right. the chat interface has its own limits separate from claude
35.
by rgbrenner 1y ago
if you work on a team most code you see isn’t yours.. ai code review is really no different than reviewing a pr… except you can edit the output easier and maybe get the author to fix it immediately
36.
by rgbrenner 1y ago
I think you misread my comment. I wasn't asking for help. I get consistent good output from Sonnet 4 using RooCode, without needing Gemini for planning. Edit: I think I know where our miscommunication is happening... The "think&qu
37.
by rgbrenner 1y ago
same. i ran a few tests ($100 worth of api calls) with opus 4 and didn’t see any difference compared to sonnet 4 other than the price. also no idea why he thinks roo is handicapped when claude code nerfs the thinking output and requires typ
38.
by rgbrenner 1y ago
Sounds similar to `ask_followup_question` in Roo
39.
by rgbrenner 1y ago
All of these reviews are irrelevant anyway because of the variations in the problems, skillset, project attributes (size, structure, etc), human variations in prompting, and a million other reasons. You should just set aside some time to tr
40.
by rgbrenner 1y ago
Recently canceled cursor. I think there's a shift happening right now with the improvements in the ability to process large context sizes and stay on task: Traditional code editing -> autocomplete -> file editing -> agent mode
41.
by rgbrenner 1y ago
my biggest complaint with it is the speed. One of the slowest agents (if not the slowest) I've tried. It's also missing quite a few features still (like checkpoints, stopping the agent mid-task to redirect, etc). But the core feat
42.
by rgbrenner 1y ago
you could just not upvote / downvote the ai answer? what do you expect those buttons to do if they don’t send anthropic the vote?
43.
by rgbrenner 1y ago
ok but what’s the crime? also english wikipedia is actually for english speakers.. so it includes countries that aren’t america. there’s a reason they didn’t name it american wikipedia.
44.
by rgbrenner 1y ago
unlike other loans sent to collections, collection costs are added to the loan. the amount varies by loan type, but it’s typically 25%. so for example, if you owe 50k: you miss a few payments, and put into default, the loan is sold.. and yo
45.
by rgbrenner 1y ago
no. junie is decent as an agent, despite it being slow (i’d put it between cursor and windsurf/copilot on quality).. but the autocomplete is anemic. they have to improve their ability to generate suggestions at all before they can star
46.
by rgbrenner 1y ago
just don’t install the plugin
47.
by rgbrenner 2y ago
that puts musks net worth at 150b. you might want to update your numbers.
48.
by rgbrenner 2y ago
This is the fallacy of composition. A party is made up of diverse independent groups who are demanding different things. When one of those groups gets "comfortable", it allows other groups to become (relatively) louder and steer t
49.
by rgbrenner 2y ago
Right now 1 out of every 5 dollars of spent GDP comes from the government. Why prefix with "Right now"? It's been roughly 20% since 1975, and only marginally lower for the 25 years before that. https://fred.stlou
50.
by rgbrenner 2y ago
"alignment" is a bs term made up to deflect blame from the overpromises the AI companies made to hype up their product to obtain their valuations.
51.
by rgbrenner 2y ago
Except Congress decides what the government will spend money on. Impoundment of appropriated funds -- what DOGE is doing -- requires the consent of congress[0]. If the Impoundment Control Act is thrown out, no institution will be safe, incl
52.
by rgbrenner 2y ago
That doesn't mention Google anywhere in it that I can find.
53.
by rgbrenner 2y ago
sounds like you have a requirement that filters out a lot of people. typically when things get rarer they get more expensive.
54.
by rgbrenner 2y ago
We cannot find qualified applicants. ... [a bunch of requirements] ... No, we don't pay as much as Meta. ... [some more requirements] The shortage on $5000 ferarris continues as well.
55.
by rgbrenner 2y ago
GPT is insufficient for AGI. It might be part of the solution one day, but it's fundamentally incapable of bridging the gap by itself. Despite what Altman/OpenAI says in public, I think he knows this too... That's why they di
56.
by rgbrenner 2y ago
"They want factual neutral reporting." If that's true, why is Fox News so much bigger than Reuters.
57.
by rgbrenner 2y ago
AMD + TSMC = ~12B vs Intel at 16B For anyone that wants a closer comparison.
58.
by rgbrenner 2y ago
I'm working on this: https://www.agiler.io Email in bio if you want to discuss specifics.
59.
by rgbrenner 2y ago
To be clear, you're not a port worker.. correct? You're just an outsider looking in, making a blanket judgement about how easy other people's jobs are; and how easy they are to replace?
60.
by rgbrenner 2y ago
microsoft doesn’t have a trademark on “ms” either. like i said, if wpengine is hoping to go into court and explain that wp is not related to wordpress, while selling wordpress services… i dont think its going to go well for them. this is go

More ›