Ted Underwood @tedunderwood.com · 26/06/2026Why does using LLMs for coding change your view of them? Bunch of things: 1. Yes, code can be verified, so they can check their work & iterate until they get it right. 2. But also: what makes coding hard is strain on short-term memory. Lots of moving parts. Long context & patience shine here. + 1112210
erols @erols.eurosky.social · 26/06/2026Writing, printing, computing and the internet: these all increased the capacity of an individual. LLMs are the latest innovations that do this. Plato on writing... He would have hated LLMs 070
Jim Clifford @jimclifford.bsky.social · 27/06/2026I think agentic LLMs are already very good at a lot of things we haven’t thought to use them for and will struggle to trust them to do. I OCR’d 7000 mostly handwritten images with the best open model (Infinity Parser 2). 120
Too Lazy to Fail @toolazytofail.bsky.social · 26/06/2026Good thread. The other thing we're noticing that some of the doubters are missing is that repeated reviews are cheap and easy, even across models, so errors compound DOWN, just like they would with a piece of code that you had a dozen people living in. 010
dnu @dnulkjkjh.bsky.social · 26/06/2026context docs changed more for me than context length. writing them forced me to externalize the mental model instead of just extending it. do you write those? 130
Marvelous Chester @aldrinbester.bsky.social · 26/06/20262 is why we absolutely hate being interrupted, it's so hard to pick up where you left off being able to offload this mental load while i keep my focus on the bigger picture is a huge win for me 050
Singularity's Bounty e/CC @catblanketflower.yuwakisa.com · 26/06/2026I'm doing some auth work and I am astounded at the standards it's pulling it. It's correcting me on security practices I've been wrong about for 20 years. So another aspect is the breadth of its vision In other areas, it's pulling in standard "best" algorithms I had no idea existed 2100
Qtain P. Faulkes @qtainpfaulkes.bsky.social · 26/06/2026Counterpoint. cat *.CodeSuffixOfChoice | grep RegexMatchingPattern Or any reasonable iteration thereof. 000
Ethan Gruber @ewg118.bsky.social · 26/06/2026Isn't #2 part of what makes coding rewarding as an intellectual exercise? 340
ftacv.bsky.social @ftacv.bsky.social · 26/06/2026I think also they're more likely to be a priori good at coding? Personally I think of them as interpolators over text, consequently the more structured and "rule-bound" the text or task (coding, poetry, translation) the better they will perform 200
Ted Underwood @tedunderwood.com · 26/06/20263. If you have multiple projects, a lot of the work is finding old files and remembering where you left off. An LLM can help here without posing as an oracle or communicative proxy: it increases *your* context length and organizational capacity. #3 will have value in lots of non-code applications! 9857
Bartek Ogryczak @vartec.bsky.social · 26/06/20261. is a bold claim, given that somewhat famously in a Turing-complete language it can’t be verified :-P 140