riku_iki

25 ms·

451.
by riku_iki 2y ago
> But almost all of the new innovation I'm familiar with in distributed systems is about training LLMs I think database space is still hot topic with many unsolved problems.
452.
by riku_iki 2y ago
No, discussion started with question "Why choose Mongo in 2025?" So, old jepsen reports are irrelevant, and most recent one from 2020 is somehow relevant.
453.
by riku_iki 2y ago
what success numbers you think are achievable for reasonably good algo-trader?
454.
by riku_iki 2y ago
Its for search/ranking functionality, not necessary GenAI.
455.
by riku_iki 2y ago
In attempt to understand your motives in this discussion, I would like to ask question: * why you are referring on 12yo reports for very early MongoDB version?
456.
by riku_iki 2y ago
they obviously didn't use vanilla postgres, but built some custom sharding on top, which is untrivial task (implementation and maintenance(resharding, failover, replication, etc)).
457.
by riku_iki 2y ago
99% of population absolutely can't build high quality LLMs or/and agents. Not talking about network effect: those who can build still need to find the way to sell those LLMs and agents. There is significant moat.
458.
by riku_iki 2y ago
> Take a look at the report for RethinkDB RethinkDB doesn't support cross document transactions, problem solved lol
459.
by riku_iki 2y ago
Are you aware: 1. That PgSQL also has issues in jepsen tests? 2. of any distributed DB which doesn't have jepsen issues? 3. It is configurable behavior for MongoDB: can it lose data and work fast, or work slower and do not lose data. T
460.
by riku_iki 2y ago
I think major providers provide PG service with cross zone availability through replication.
461.
by riku_iki 2y ago
I think there is no distributed db on the market available with features parity to PgSQL. Distributed systems are hard, and sacrifices need to be made.
462.
by riku_iki 2y ago
things still can be true, even if being wrapped into meme videos by haters..
463.
by riku_iki 2y ago
Mongo is real distributed and scalable DB, while postgres is single server DB, so main consideration could be if you need to scale beyond single server.
464.
by riku_iki 2y ago
I think Meta folks just don't know how to come to this market and build something potentially profitable, and doing random stuff, because need to report some results to management.
465.
by riku_iki 2y ago
> for models that themselves take many weeks if not longer to train. they all have foundational heavy-trained model, and then they can do follow up experimental training much faster.
466.
by riku_iki 2y ago
that's because stacks/apis/ecosystems are super complicated and require lots of reading/searching to figure out how make things happen. Now this time will be reduced dramatically and devs time will shift on more novel th
467.
by riku_iki 2y ago
Is there plans to add websearch function over some core websites (SO, API docs)? Competitors have it, and in my experience this provide very good grounding for coding tasks (way less API functions hallucinated).
468.
by riku_iki 2y ago
> The wealthy cannot maintain control over this technology. Again, there is no moat. in a meantime, gap in wealth is keeping increasing, and AI will be one another multiplier.
469.
by riku_iki 2y ago
History absolutely shows that multiple empires where power/wealth was concentrated in hands of few people sustained for hundreds years. Revolts could be successful or not successful, with tech advancements in suppression (large scale s
470.
by riku_iki 2y ago
why do you think it will lead to collapse exactly?
471.
by riku_iki 2y ago
It depends. If there will be one/few winners on the market, they will dictate price after human labor out-competed through the price or quality.
472.
by riku_iki 2y ago
> people will pay an AI lawyer $500 to write a document instead of paying a human lawyer $500 to write a document. there will be caste of high-tech lawyers very soon which will be able to handle many times more volume of work thanks to A
473.
by riku_iki 2y ago
poor got more comfortable life because of globalization: they became useful labor for corps. Things will go back to previous state if their jobs will go to AI/robots.
474.
by riku_iki 2y ago
consumer demand will shift from middle-class demand (medium houses, family cars) to super-rich demand (large luxury castles, personal jets and yachts, high-profile entertainment, etc) + provide security to superrich (private automated polic
475.
by riku_iki 2y ago
> Maybe LLMs can't work on huge code bases yet its also not just about code base size, but also about your expectation of output quality/correctness.
476.
by riku_iki 2y ago
> I'm pretty skeptical of that 75% on GPQA Diamond for a non-reasoning model. could that benchmark be simply leaked to training data as many others?
477.
by riku_iki 2y ago
I think Sonnet doesn't have web search integrated, and I suspect because of this I receive more hallucinated lib APIs compared to gpt.
478.
by riku_iki 2y ago
paper says they created e2e tests to check if task completed successfully.
479.
by riku_iki 2y ago
There are llms below 1000 in the leaderboard
480.
by riku_iki 2y ago
And DeepSeek is just 3% behind. It seems in that benchmark all LLMs perform well and top is formed within some statistical error.

More ›