riku_iki

15 ms·

301.
by riku_iki 1y ago
> your position is that the VC does not suffer any significant consequences? it is likely they won't suffer any significant consequences
302.
by riku_iki 1y ago
> that a prospective $2B investor has access to more information about the company's plans than we do and that their investment is at least some indication that there is a real product in the pipeline. there is a funnel of investors
303.
by riku_iki 1y ago
> but presumably the VCs have some idea of what Thinking Machines is building you imply that VCs are rational because bet their own money, which in current complicated world probably is not true. VC funds get money from complicated funne
304.
by riku_iki 1y ago
> I have 100+TB datasets and with a large enough SSD/RAM for L1/L2 arc ZFS edges out. and your claim is that you tested it against btrfs on the same workload? Maybe you could post some specific numbers from running command from
305.
by riku_iki 1y ago
I test on 2TB datasets. Do you have any specific pointers which would support your claim?
306.
by riku_iki 1y ago
actually, my personal benchmarks and multiple accounts in internet say it is much slower than btrfs under the load.
307.
by riku_iki 1y ago
I would recommend btrfs as general purpose root filesystem. Any FS will have people encountering data loss. I can believe btrfs has N times higher chance of data loss because its packed with features and need to maintain various complicated
308.
by riku_iki 1y ago
Ok, I know this, not sure what is your point.
309.
by riku_iki 1y ago
> but as you may guess we did not use btrfs ever again. there are scenarious where btrfs is currently can't be replaced: high performance + data compression.
310.
by riku_iki 1y ago
> But being unable to mount, is a bug that makes a filesystem untrustworthy and useless. we are in disagreement on this. If partition table entry corrupted, you can't mount without some low level surgery. > And how would RAID he
311.
by riku_iki 1y ago
> Well, maybe it was my fault but since I'm using the exact same system with the same hardware right now (same NVMe SSD), I really doubt that. anecdotes could be exchanged in both directions: I run heavy data processing with max pos
312.
by riku_iki 1y ago
> One day, the system froze and after reboot the partition was unrecoverably gone (the whole story[1]). it looks like you didn't use raid, so any FS could fail in case of disk corruption.
313.
by riku_iki 1y ago
> However, it does help to contrast the much more common, "use Postgres for everything" type sentiment. I think sentiment is to use "for everything in 99% business cases", which involves few 100GB of data with some th
314.
by riku_iki 1y ago
> the real question is why other LLMs didn't do as well in this benchmark. they do. There is a cycle for each major model: - release new model(Gemini/ChatGPT/Grock N) which beats all current benchmarks - some new benchmar
315.
by riku_iki 1y ago
> For example, here's a list of GenAI projects at Capital One that involve RAG/Vector databases: https://sumble.com/l/6sDqKmhyAH requires you to sign in, which then follows by marketing emails.
316.
by riku_iki 1y ago
but you can't build 5B startup in 10 months with plain old coding..
317.
by riku_iki 1y ago
would be great if I could provide prompt/context, for example specifying domain and scope instead of getting random proposals.
318.
by riku_iki 1y ago
you also could compare internal dimensions between 3: 2005 accord vs 2025 civic vs 2025 accord, I kinda believe 2025 civic will be closer to 2005 accord. But I mostly referred on class in general: 2005 accord was basic appliance economy car
319.
by riku_iki 1y ago
2005 accord is a 2025 civic class wise
320.
by riku_iki 1y ago
Many humans can't be trusted with important work too, that's why we have all job interviewing and performance review processes, which are messy, costly and inefficient. With AI, companies can built some rigid analytics/tests&
321.
by riku_iki 1y ago
Iran and allies already did what they could during Gaza escalation. Their projection power is rather limited.
322.
by riku_iki 1y ago
And how was result? Did you verify that it found reliable source of data?
323.
by riku_iki 1y ago
those who try to fix them are fighting alone against huge corps which try to abuse them..
324.
by riku_iki 1y ago
For inference cost, I don't see how this is different from cloud providers vs dedicated server providers, where AWS is 5-10x more expensive than hetzner. Somehow cloud providers manage to add lots of extra-cost on offering.
325.
by riku_iki 1y ago
> The company and product doesn't matter and they'll happily burn it to the ground if that makes more money. they do not burn it to the ground, community has open source high quality product for absolutely free developed on inv
326.
by riku_iki 1y ago
I think LLM will also hallucinate way more if you will give confirmation: "oh, this is so true, could you continue and expand more?". Its the issue of reward function, both humans and LLMs are trained on pleasing clients as one of
327.
by riku_iki 1y ago
He is likely just not very smart: "Law enforcement sources said they turned up multiple Polaroid pictures of the victim being tied up and tortured"
328.
by riku_iki 1y ago
so, its just a habit now..
329.
by riku_iki 1y ago
> You can see the results and links to our code in the write-up here: https://surrealdb.com/blog/beginning-our-benchmarking-journe ... page says your benchmark runs on 5M of records only. Is it incredibly small datas
330.
by riku_iki 1y ago
How scalable is your DB in your tests? Could it be performent on graphs with 1B/10B/100B connections?

More ›