riku_iki

12 ms·

1. 181.▲
   by riku_iki 1y ago

   > reinforcement learning, which is what most people mean when they talk about reasoning LLMs popularity contest has no place in tech discussion, and even then not clear on which evidence you make such statement. imo, reasoning model is
2. 182.▲
   by riku_iki 1y ago

   so, give ref on results which prove that they reliably can multiply arbitrary numbers > Maybe at least give a cursory glance at a paper before trying to cite it to support your invalid point? they use CoT aka reasoning steps
3. 183.▲
   by riku_iki 1y ago

   I think you just made lots of handwaving statements. Here is result which says LLMs can't do multi-digit multiplications well: https://arxiv.org/pdf/2510.00184
4. 184.▲
   by riku_iki 1y ago

   > then reasoning LLMs seem to pass that bar as they do very well programming and maths competitions. it could be this is just result of good stochastic parroting and not reasoning. Both of those niches are narrow with high amount of tr
5. 185.▲
   by riku_iki 1y ago

   cassandra doesn't have ACID, so you will start dealing with tons of other problems.
6. 186.▲
   by riku_iki 1y ago

   I skimmed first page of google search results, and couldn't find such statements.
7. 187.▲
   by riku_iki 1y ago

   > they are the ones with the massive VO2 maxes my intuition says that interval running is the best regime for this, and most time efficient (like 15 mins per session: 6x30sec sprints + 6x2m jogging)
8. 188.▲
   by riku_iki 1y ago

   Could you run some query like select sum(banch of columns) from my_table and see how long it will take? 600k rows is likely less than 1GB of data, and should take about second to load into RAM on modern nvme ssd raids.
9. 189.▲
   by riku_iki 1y ago

   You still looking at small part of the picture. Big part of the picture is confrontation of Western World and allies against China, Russia, Iran. The winning strategy could be to weaken enemy now: strike on Russia military/oil infra, a
10. 190.▲
    by riku_iki 1y ago

    Or prevent WW3. Hitler could be stopped in Czechoslovakia.
11. 191.▲
    by riku_iki 1y ago

    > AI in medicine is challenging because AI is bad at systems thinking, citation of fact and data privacy. main question is if humans are better than that. I have experiences with doctor: he gave prescription of Xmg, I am asking why, he s
12. 192.▲
    by riku_iki 1y ago

    Google grew partially because got weak penalties in anti monopoly case
13. 193.▲
    by riku_iki 1y ago

    And do you know what was the transaction+replication story there? Were all changes transactionally and synchronously replicated?
14. 194.▲
    by riku_iki 1y ago

    > MyISAM engine MyISAM didn't support ACID transactions, thus can't be apples to apples comparison, its just very different niches. There are plenty of distributed databases on the market today which could be used if you don&#x
15. 195.▲
    by riku_iki 1y ago

    > I've seen multi-master HA architecture work out great for 10M+ DAU games could you tell what kind of DB was that so we can understand if it is apple to apple comparison to multi-master PG?
16. 196.▲
    by riku_iki 1y ago

    > Write throughput: ~1.88M records/sec (MessagePack, untuned) this doesn't sound like much, unless records are very large..
17. 197.▲
    by riku_iki 1y ago

    what about marketing and sales?
18. 198.▲
    by riku_iki 1y ago

    they don't care much: https://www.selc.org/press-release/musks-xai-explores-anothe...
19. 199.▲
    by riku_iki 1y ago

    I suspect it will mostly be fossil power capacity, which is much easier to scale up
20. 200.▲
    by riku_iki 1y ago

    vps provides some advantages, for example snapshoting
21. 201.▲
    by riku_iki 1y ago

    > Which also illustrate that a stable business does not imply a stable product. there is no absolutely stable product, some weird stuff can happen to everything, its question of probabilities, what the chance that some underfunded startu
22. 202.▲
    by riku_iki 1y ago

    that list is catchy, but if you dig into details, there are many caveats, for example many product were not killed, but replaced with low friction, for example hangout, many products were created at period of time when there was that google
23. 203.▲
    by riku_iki 1y ago

    > elbows flaring out instead of being held close to the body not sure about others, but I think this one is already controversial, having elbows close to the body shifts load to triceps. Canonical form is to have arms 45 degrees to body
24. 204.▲
    by riku_iki 1y ago

    > Why is it okay to just accept that the hyperscalers are 100% closed, but the minute a smaller player tries to play "open-ish" with "fair source" we crucify them? besides morale, from pure business point of view, hyp
25. 205.▲
    by riku_iki 1y ago

    hypothetically LLM absorbed lots of world knowledge, and it can trace lots of deep correlations between various factors.
26. 206.▲
    by riku_iki 1y ago

    they all were not part of hyped roadmap with XXB of investments and XXXB evaluation. Throwing tons of money can often make things happen.
27. 207.▲
    by riku_iki 1y ago

    I read this in wiki too, but couldn't trace where this came from.
28. 208.▲
    by riku_iki 1y ago

    any reliable record for this?
29. 209.▲
    by riku_iki 1y ago

    > Btrfs is also a clusterfuck. Perfect example for the consequences of NIH-syndrome. building fs with such large features set is just untrivial task, and btrfs one of very few who made it, so it is absolute success story.
30. 210.▲
    by riku_iki 1y ago

    I am sure he can ask some frontier model: hey, find me justification and reputable references for statement X, do research as deep as required to come with results.