Husband, father and aspiring digital carpenter and investor. Co-founder and CTO at Elvex. Formerly Koru, @parsely, Globe and Mail and Rogers Media. 🇨🇦
Toronto, ON, Canada
Joined November 2008
- Tweets5.5K
- Following1.4K
- Followers797
- Likes13.1K
Mike Sukmanowsky 🇺🇦 retweeted
This is notable: Opus 4.5 is ~60% more expensive than Sonnet ($25/million output compared to $15/million) but if it can use 76% fewer output reasoning tokens for the same complex task it may end up cheaper!
Mike Sukmanowsky 🇺🇦 retweeted
we've released a CRDT extension for Postgres (experimental and open source)
why? A CRDT is a special data type used for collaboration - think of a Google Doc. The changes are merged even as people are typing simultaneously
these CRDTs are then serialized on a server and then stored in a database.
with a Postgres extension, we give the database a more "native" experience - they can be merged on the database, indexed, and searched.
In the future you should be able to do something like this:
create table blog_posts (
id uuid primary key,
title text,
body crdt default '{}'
);
then you can post the changes to the database and it will merge automatically (rather than overwriting eachother's changes)
you can use websockets (like our Realtime server) for very rapid changes between users, batching the changes to your database
Mike Sukmanowsky 🇺🇦 retweeted
Observing some people close to me with chronic health conditions, it's striking how useful Reddit frequently ends up being. I think a core reason is because trials aren’t run for a lot of things, and Reddit provides a kind of emergent intelligence that sits between that which any single physician can marshal and the full rigor of clinical trials.
Why aren’t trials run for a lot of things? Well, they’re of course slow and expensive (median cost of $19M for a pivotal trial in 2015[1]; after adjusting for inflation and other phases, maybe that corresponds to a total of $40M today?). But they’re also hard to fund when the intervention in question lacks IP protection since the ensuing knowledge can’t be monetized. As such, trials for diet, over-the-counter supplements, and lifestyle interventions are under-pursued. To give one prosaic example, lots of people think that magnesium improves sleep, but, as far as I know, no trial has ever been run assessing its ability to improve sleep in non-elderly adults without sleep disorders.
So, Reddit — in a pretty unstructured way — makes a limited kind of “compounding knowledge” possible. Best practices can be noticed and can imperfectly start to accumulate. For people with chronic health problems, this is a big deal, and I’ve heard lots of stories between “I found something that made my condition much more manageable” all the way to “I found a permanent cure in a weird comment buried deep in a thread”. (Of course, one also sees this outside of medical conditions. I’ve enjoyed the recommended routine in the BodyWeightFitness subreddit, as a comparable kind of distilled practical wisdom[2].)
An interesting and somewhat more formalized example of this approach was recently used for long COVID and published earlier this year[3]. After surveying 3,900 individuals, the paper analyzes patient-reported outcomes for 150 different treatments, yielding the figure reproduced below. There are evidently no silver bullets, but it is striking that, say, about half of people find that antihistamines are helpful. I know a number of people who found the learnings from this study to be impactful in improving their daily quality-of-life.
Seeing this paper and the Reddit experience makes me wonder whether the approach could somehow be scaled: is there a kind of observational, self-reported clinical trial that could sit between Reddit and these manual approaches? Should there be a platform that covers all major chronic conditions, administers ongoing surveys, and tracks longitudinal outcomes?
I don’t really know what the best way to go about this would be, but it feels to me that there could be something important here. There’s a lot of latent data in patients’ subjective experiences that is not today being properly gathered or analyzed.
Mike Sukmanowsky 🇺🇦 retweeted
Ok, @OpenAI GPT-4o can generate 2D character animation sprite sheets for your vibe coded video game
Prompt below 👇
Mike Sukmanowsky 🇺🇦 retweeted
Which of America's bold new cohort of Republican leaders is bold enough to answer this?
Mike Sukmanowsky 🇺🇦 retweeted
defending freedom and not going anywhere.
thank you to everyone who is standing with us.
To take a tiny bit of air out of the DeepSeek balloon, here's DeepSeek R1 8b falling into an endless thinking rabbit hole gist.github.com/msukmanowsky…
Mike Sukmanowsky 🇺🇦 retweeted
I understand that k8s and containers is impossible to avoid these days, but when shit goes wrong, the layers of the onion you have to unravel are wild. Gone are the days you quickly create a coredump or attach gdb/strace etc. Now debugging takes ages and sometimes is not helpful.
Don't sleep on developer experience!
Mike Sukmanowsky 🇺🇦 retweeted
Big announcement today @huggingface: We now have a unified API for tool use across models from @MistralAI, @AIatMeta, @cohere, @NousResearch and more!
That means that you can reuse the same simplified, portable code to add tool capabilities to all of those models! 🧵
Mike Sukmanowsky 🇺🇦 retweeted
I think the take-away message from the latest LLM results is how few deep secrets the big AI labs actually had. GPT-4 felt like magic in 2023, but one year later, we now have multiple open models with better outputs using less than a tenth of the parameters.
Mike Sukmanowsky 🇺🇦 retweeted
I got 12 tokens/second out of Mixtral-8x7B with NO GPU - more than fast enough for live chat! You can too!
Hardware:
Supermicro MBD-H13SSL-N
AMD EPYC 9124
12 x 16GB 4800mhz DDR5 ECC RDIMM
Software:
llama.cpp + Mixtral Q8 (on @huggingface)
For why this works, thread below 🧵
Mike Sukmanowsky 🇺🇦 retweeted
Today in timezone madness: Lebanon had a last minute announcement of DST postponement. The Olson database complied. Now there is a dispute if it happens, and *parts* of the government reversed it. The country now has two timezones concurrently depending on who you ask.
Alright, I'm out.
#GoodByeTwitter
