kitallis

kitallis

@repartee.party

๐Ÿ–‡๏ธ kitallis.in ๐Ÿ’ฌ https://signal.me/#eu/1MJzVaEQabfP9zO4YQ5t3Y6miP-0oYtlNMo4OYwWNiQeW5voWGFD5ZWopF9Sf8Mu

35 followers61 following67 posts

the hilariously ironic thing about Ather crowdsourcing bumpy roads and potholes throughout the city is that in Bangalore, you now get alerted literally every minute to the point that it's completely useless to have that turned on

decided to emit perfetto traces for instrumenting an agent, and it works... surprisingly ok. dump reasoning tokens, prompts, counters into args and run queries directly on ui.perfetto.dev, no Langfuse et al to setup

kitallis reposted

Replying to @masnick.com

And it says *100% AI written* with "high confidence." It is 100% wrong. I say... AI detector tools are about as reliable as AI writing tools.

a lot of my feed in various places lately is people pretending to know how LLMs and agents are not going to actually replace certain parts of my software engineering job and taste (and related subjective functions) is the real moat or how I'm not using the full power of these tools yet

I need a new/better word for code that's hand-crafted but with a high-degree of AI assistance. Maybe: ๐˜ฑ๐˜ณ๐˜ฐ๐˜ฅ๐˜ถ๐˜ค๐˜ฆ๐˜ฅ ๐˜ฃ๐˜บ ๐˜ฌ๐˜ช๐˜ต๐˜ข๐˜ญ๐˜ญ๐˜ช๐˜ด

the real super intelligence is instantly spotting that the Nikhil Kamath and Darioโ€™s interview is Four Seasons in Bangalore because of the specific carpet pattern and background landscape

I started watching The Leftovers during COVID, and I was so pulled in by it that I decided to NOT watch the last 3 eps indefinitely, so that I can come back to it 10-20y later and hopefully feel like that again (5y and counting).

Astreia@redoubters.bsky.social

Given that apparently the Stranger Things finale was meh (idk, didn't watch it, just the scuttlebutt) and we're not that far removed from the disastrous GOT finale that retroactively made everyone have never cared about the show: What's the *best* ending to a show you've ever seen? Quote/reply etc

I wonder if we can do llm swe benchmarks that specifically decontaminate certain "ergonomically human" ideas from training, like tasteful, dev-ex oriented libraries or frameworks (say rails) and see if we can progressively prompt (+elicit) our way to something that everyone "enjoys" using

i have seen like 4 full-length sam altman interviews in the past months (including the nikhil kamath one) and there's something very weirdly unsophisticated about his pitch about AI every time it always slants towards something like start a company and write code or plan itineraries and do science

like we know people are doing a lot more with these things, but he just sort of glosses over, and it's unclear whether it's intentional or he just doesn't have the words to convey specificity

so am I getting this right, to get openai api credits: - you have to prepay - the amount is only valid for a year - you can't extend it - they won't refund excess - funds aren't transferable

the trash segregation game at #rubykaigi is intense and awesome, 5+1 ways and you can't weasel your way out because there are people monitoring your droppage and will force you to do it right

there was a time in the last couple of years when starting up where I was like linkedin is just a subtle perspective shift, there is obviously grift, but people are trying to make things happen in a direction they want to pursue and this is generally a good thing, but now I'm like this is all grift

it's also obviously meta-grift, most people know what they are doing, and it's a youtube thumbnail tactic thing but when you do too much of it becomes just like stuffing momos inside pizzas and burgers and pasta and noodles and even inside the tandoor

kitallis reposted