tibbar

10 ms·

151.
Why are you telling me this?(jacobbrazeal.wordpress.com)
2 points by tibbar 1y ago|0 comments
152.
by tibbar 1y ago
I think a spiritually similar assignment would be: implement multiplayer Google Docs using a CRDT, server, and clients, supporting a certain class of documents that are shown to you during the interview. And imagine that the CRDTs are not d
153.
by tibbar 1y ago
Oh, what I meant was, this was an interview with a company that produces a coding assistant. Ironically, you really weren't supposed to use an LLM much for the interview, although I think you could use the company's product if you
154.
by tibbar 1y ago
That is fair. I suppose what I meant is, the idea of working at one of these companies can be really exciting, almost a fantasy, but in practice: it might actually hurt you in many ways. 'Look what they make you give', as a cert
155.
by tibbar 1y ago
I recently did a round of interviews at various AI companies, including model labs, coding assistants, and data vendors. My first takeaway is that, wow! the interviews are very hard, and the bar is high. Second, these companies are all sele
156.
by tibbar 1y ago
Not OP, but I think TLM works best when it's a transitional role. You have someone you want to groom into a full-time manager, and you have a team that you plan to grow over time. TLM itself is not that efficient, but can lead to stron
157.
by tibbar 1y ago
For a team that size, you would assume the manager is only spending around half of their time on people management and probably around half their time working directly on whatever the team does. It can be a good arrangement if the goal is
158.
by tibbar 1y ago
Yet I think most people err in the other direction. They 'know' the basics of health, of discipline, of charity, but have a hard time following through. 'Take a simple idea, and take it seriously': a favorite aphorism o
159.
by tibbar 1y ago
As you say, generics really only apply to typed languages, and they help solve very legitimate annoyance in most of those languages -- you often have some library helper or algorithm that can apply to a wide range of things, but those thing
160.
by tibbar 1y ago
Really? I thought the analogy was pretty good. Here senses refer to how the machines perceive text, IE as tokens that don't correspond 1:1 to letters. If you prefer a tighter comparison, suppose you ask an English speaker how many vowe
161.
by tibbar 1y ago
Are you saying that the hardware OpenAI used for inference on previous models is incompatible with the hardware used for GPT-5? Or are you perhaps saying that GPT-5 is just cheaper to run than the old models? > It would be stupendously e
162.
by tibbar 1y ago
With major redesigns, you often can’t directly compare the two versions —- they are different enough that you actually want people to use them in a different way. So it’s not that the new version is “worse”, it’s just different, and it’s po
163.
by tibbar 1y ago
It's a good reminder that OpenAI isn't incentivized to have users spend a lot of time on their platform. Yes, they want people to be engaged and keep their subscription, but better if they can answer a question in few turns rather
164.
by tibbar 1y ago
While it's possible that the LLMs are intentionally throttled to save costs, I would also keep in mind that LLMs are now being optimized for new kinds of workflows, like long-running agents making tool calls. It's not hard to imag
165.
by tibbar 1y ago
I've worked on many migrations of things from vX to vX + 1, and there's always a tension between maximum backwards-compatibility, supporting every theoretical existing use-case, and just "flipping the switch" to move eve
166.
by tibbar 1y ago
You can memorize the correct opening moves in chess. For maybe my first year playing chess, I just YOLO'd the opening moves. My judgement there was probably not much worse than the rest of my play, but with other players playing engine
167.
by tibbar 1y ago
It's honestly impressive. Even during the major tech cuts of 2022-2023, I think many companies ended up about the same size or a even little larger over the course of a year, due to all the other hiring happening and the fact that the
168.
by tibbar 1y ago
The luckiest thing I ever did was probably burning out, quitting my small-town engineering job, and then, after a few months, getting in at an AI-related SF startup before AI really took off, during that bubble when it was a little easier t
169.
by tibbar 1y ago
This depends on the type of work being done. Sometimes the cost of verification is much lower than the cost of doing the work, sometimes it's about the same, and sometimes it's much more. Here's some recent discussion [0] [0]
170.
by tibbar 1y ago
I, too, walk funny. I distinctly remember an authority figure trying to teach me how to walk properly as an adolescent, and other people in my life occasionally comment on it. I guess I walk by swinging my legs from the top and placing my e
171.
by tibbar 1y ago
Well, since he worked on a brand-new product, it probably didn't matter too much.
172.
by tibbar 1y ago
These are funny, but I do note that they start recycling. I’m curious if these are actually being generated or come from a stock library?
173.
by tibbar 1y ago
Until recently, training-time compute was the dominant cost, so we're really just getting started down the test-time scaling road.
174.
by tibbar 1y ago
Interesting. I'd guess this technique should probably work with any SOTA model in an agentic tool loop. Fun!
175.
by tibbar 1y ago
I mean, either they cheated on evals ala Llama4, or they have a paradigm that's currently best in class in at least a few standard evals. Both alternatives are possible, I suppose.
176.
by tibbar 1y ago
The trick they announce for Grok Heavy is running multiple agents in parallel and then having them compare results at the end, with impressive benchmarks across the board. This is a neat idea! Expensive and slow, but it tracks as a logical
177.
Baseball, Pitcher Wins, and Life(jacobbrazeal.wordpress.com)
3 points by tibbar 1y ago|0 comments
178.
How to build internal tools(jacobbrazeal.wordpress.com)
2 points by tibbar 1y ago|0 comments
179.
by tibbar 1y ago
Shareholders retain their stock, but also received a dividend equivalent to selling all their shares at a premium to the previous valuation. It is actually an incredible deal for them. (Source: I'm one of them.)
180.
by tibbar 1y ago
Internal research assistants. Essentially 'deep research' hooked up to the internal data lake, knowledge bases, etc. It takes some iterations to make a tool like this actually effective, but once you've fixed the top N common

More ›