On Microsoft Employees dealing with Github Copilot PR - May 2025

    Related links: Reddit thread, HN thread

    A few thoughts:

    I fully understand the schadenfreude on the part of the SWE community. These things are possibly an existential threat to their livelihoods and the negativity is fully understandable if not misguided, but these victory laps seem to be repeating a common pattern of not waiting a couple months(!) before these systems improve, very publicly, from a mix of better models and now actively incremented training data from MSFT employees. This is the worst this system will ever be and looking in the improvement rate of the last few months should teach people to wait a little bit before condemning the effort fruitless

    On a tangential note, I wonder if this public effort is somehow related to Github's new updated rate limits (hn discussion). All this juicy data PLUS expert human feedback out in the open ready to be used for distillation is an obvious target to upstarts trying to leapfrog capability development

Comments

Post a Comment

Popular posts from this blog

JIT and GIL removal are not even my most anticipated Python 3.13 features

     Fun story: I came to Python through data analysis - so I wasn't a "python/programming expert" - and for an unreasonably long time used exclusively csv, parquet (thanks to pandas) and later JSON to persist data from my projects (local projects, of course. Mongo, RDS and Athena were always there). At work this was 100% enough (mostly parquet), but it always felt a little clunky to use JSON for a quick cache, especially when said cache grew larger for longer running processes or longer projects. Opening, parsing then serializing back again becomes slow surprisingly fast.      Then after a few years I decided to start using SQLite. I knew it existed, but always felt like it was a bit of a hassle to define a schema and "get" and "set" functions in all projects just to have a basic persistent k-v functionality. It was OK and had the extra benefit of allowing me to safely check this cache on a different jupyter notebook process when running longer proce...
Read more

An interview with Steve Wozniak by Jessica Livingston cured my AI anxiety

Note: there's no hidden knowledge contained in this post, it's just a reflection on how reading someone's reaction to an analogous situation changed my outlook Like everyone who works in an information job, I have been worried these last few months about how AI would impact my place in the labor force in the coming decade. Even though I didn't get to the depressed extremes we've been seeing on the Internet, I could not shake off the feeling that what I could offer to a company - no, to the whole economy, which just means "other people" - would get a lot less valuable. Come mid January, I wast taking a week off and had taken to read a few books since I read so little non-technical content on a day to day basis. One of these books was "Founders at work" by author and YC co-founder Jessica Livingston. The book is a series of interviews conducted by Jessica with the founders of some of the most important tech companies - sometimes specific products -...
Read more

What if regular exercise is the best cognitive exercise?

File this under highly speculative. I’m sure there is better research on this topic than this writing This past week, for no obvious reason, I considered a few thoughts at the same time: Exercise is supposedly one of the best anti-aging “drugs”, and also highly effective (relatively speaking) against cognitive decline There’s this vague idea that after you become “professionally inactive”, you accelerate your cognitive decline To avoid cognitive decline, one must keep “exercising the brain”, so the incentive for seniors to have hobbies, do crosswords, sudoku, etc Now, from this point it seems that I’d go to a very commonplace conclusion about how exercising is probably better than sudoku or something, and to be frank that’s exactly where I’m going. But another two things crossed my mind at the same time: Recent advances in AI, and less recent advances in computing in general showed that a lot of the things we consider “cognitively hard” - from symbolic manipulation to arithmetic, “...
Read more