riku_iki

16 ms·

331.
by riku_iki 1y ago
github also mentions it has pentagonal shape..
332.
by riku_iki 1y ago
My point is that in certain domains your regular devs can't solve problem, there are some 10x experts who can solve it. For some problems, one 10x expert could be enough, for others you need some team (small), and they will be able to
333.
by riku_iki 1y ago
> Think of it this way: you can either manage 5 people that solve a problem there are levels of problem you will struggle to find 5 people to solve it on avg senior dev salary > Not in "big tech". especially in big tech, wh
334.
by riku_iki 1y ago
> Less hiring means fewer natural opportunities for growth. IC's natural way of growth is to produce larger impact by solving harder problems, there are always hard unsolved problems which hold some business opportunities.
335.
by riku_iki 1y ago
Actually, most researchers of transformer paper founded various startups.
336.
by riku_iki 1y ago
> Without the transformer architecture they invented by funding basic research, there would be no modern LLMs. one could argue that transformers are nothing without attention layer, which was not invented at google.
337.
by riku_iki 1y ago
Not sure how they could hold lead in case they lose search traffic.
338.
by riku_iki 1y ago
> It's such a ridiculous proposal that would completely destroy Google's business. it won't. My bet is that bing and some other indexes are 95% Ok for average Joe. But relevance ranking is much tougher problem, and "g
339.
by riku_iki 1y ago
> they only communicate when one node has exclusive ownership over a block another node needs to write to and it needs to communicate that he has exclusive ownership, and also after each writes you need to invalidate cached data on other
340.
by riku_iki 1y ago
PG ecosystem actually has multiple shared nothing cluster implementations, example is www.citusdata.com, unlike RAC, where masters need to sync to accept writes, so technically write load is not distributted across servers.
341.
by riku_iki 1y ago
> Postgres calls a server is a multi-process instance on a single machine, often a commodity cloud VM. What ExaData calls a server is a dedicated bare metal machine with at least half a terabyte of RAM and a couple hundred CPU cores, wit
342.
by riku_iki 1y ago
> 2TB of data, 4TB of backup storage and peak of 64 cores with elastic scaling it's about 6k/month right, but foundationdb and similar is for cases where you actually need throughput of hundreds servers, for what you described
343.
by riku_iki 1y ago
> few hundred nodes at most, and Oracle RAC/ExaData clusters can scale up that far too and how much license fee it will cost?
344.
by riku_iki 1y ago
> industries that are less legible on the open internet I think there will be also wave of private LLMs fine-tuned on corporate data, and it will be also good tools.
345.
by riku_iki 1y ago
I didn't say I blindly trust generated code, it needs to be read through and covered by tests, but tasks like generate boilerplate code for some lib I never used before and which would take time to locate and read through documentation
346.
by riku_iki 1y ago
I (and many others) use LLM for coding tasks multiple times a day, its very unlikely they will go the way of NFTs.
347.
by riku_iki 1y ago
I am looking from perspective of vibe coding task for some regular dev. If he has access to github from his current IDE, nothing prevents Cursor from having the same level of access.
348.
by riku_iki 1y ago
there is no lock in on github, since communication happens through git protocol, so no external tool/client can be locked out.
349.
by riku_iki 1y ago
they have userbase, revenue, super-fast growth, leading market position and capable team.
350.
by riku_iki 1y ago
> Results with this method are significantly more limited compared to all the power open-weight finetuning gives you (and the skillset needed in return). I am not sure I understand why you are so certain that finetuned top market models,
351.
by riku_iki 1y ago
> But one cannot deny RAG is a lot more accessible to teams which are likely developers / AI engineers compared to ML engineers/researchers. It looks like major vendors provide simple API for fine-tuning, so you don't need
352.
by riku_iki 1y ago
> RAG is infinitely more accessible and cheaper than finetuning. it depends on your data access pattern. If some text goes through LLM input many times, it is more efficient for LLM to be finetuned on it once.
353.
by riku_iki 1y ago
> I partly disagree, software still sucks and it's a great time to build. you can build great software, but the bigger challenge is to defeat network effect of preoccupied/monopolized market.
354.
by riku_iki 1y ago
they want speed check in California, which is major market, and automakers likely will comply.
355.
by riku_iki 2y ago
Its easy to bet, more and more people switching to chatbots with tasks they previously used search for, and this can dramatically affect Google main revenue stream: Search Ads.
356.
by riku_iki 2y ago
Its just it is not easy to come into specific pre-occupied space for outsiders.
357.
by riku_iki 2y ago
But clickbench doesn't have joins..
358.
by riku_iki 2y ago
interesting advice: monitor how often ClickHouse crashes with segfaults. Doesn't sound like production grade quality product.
359.
by riku_iki 2y ago
maybe one line of defense is that Chrome is good for society, but no one wants to buy and support it..
360.
by riku_iki 2y ago
Solution could be to mix RL training with foundational knowledge training, so LLM can refresh memory and not forget things.

More ›