tucnak

25 ms·

331.
by tucnak 1y ago
> I'm just curious - now much of your portfolio is AMD I'm always just like... who are you people: financiers, or hackers? :-) I don't work for TT, but I am a founder in the vertical AI space. Firstly, every major player i
332.
by tucnak 1y ago
The actual lead times on similarly-capable GPU systems are so long, by the time your order is executed, you're already losing money. Even assuming perfect utilization, and perfect after-market conditions—you won't be making any mo
333.
by tucnak 1y ago
What a load of nonsense. MI210 effectively hit the market in 2023, similarly to H100. We're talking about datacenter-grade, two-year out of date card, and it's already "ancient history?" No wonder nobody on this site tru
334.
by tucnak 1y ago
Unfortunately, GPU's are old news now. When it comes to perf/watt/dollar, TPU's are substantially ahead for both training and inference. There's a sparsity disadvantage with the trailing-edge TPU devices such as v4
335.
by tucnak 1y ago
This misconception is repeated time and time again; software support of their datacenter-grade hardware is just as bad. I've had the displeasure of using MI50, MI100 (a lot), MI210 (very briefly.) All three are supposedly enterprise-gr
336.
by tucnak 1y ago
> Yes, the savings in labour translates to savings in time. This should be obvious. Under some circumstances; arguably, in only a handful of circumstances. The colloquial 10xer may as well as get the job done at a fraction of the labour
337.
by tucnak 1y ago
*s/major/minor
338.
by tucnak 1y ago
The proposition that Electron apps are somehow "saving time" is preposterous! Of course, the real cost-saving is in labour—web development presents a radically lower barrier to entry compared to even non-native, cross-platform UI&
339.
by tucnak 1y ago
Precursor just happens to use Xilinx Spartan 7-class, a 100 MHz VexRISC-V, RV32IMAC + MMU, 4k L1 I/D cache. But these days you could buy Versal devices with 1M+ LUT, and up to 100K DSP slices, in under a thou. However, integrating it—w
340.
by tucnak 1y ago
Or best buy something like https://www.crowdsupply.com/sutajio-kosagi/precursor or some other FPGA-based platform to retain the programmable logic capability, you never know whether you're going to need it, and sh
341.
by tucnak 1y ago
This is referring to implementation. I argued, simply augmenting the API to match grammar property of the llama.cpp library was not enough, as it doesn't recognise the reality of Ollama consumption. None of the existing Ollama clients
342.
by tucnak 1y ago
Your example demonstrates how little of PostgreSQL syntax you actually know.
343.
by tucnak 1y ago
A PR that nobody would accept? There were like 20 pull requests implementing grammar sampling parameter handling (largely distasteful) at the time, over 16 months, absolutely zero acknowledgement from ollama maintainers. Sending patches to
344.
by tucnak 1y ago
The laborious extents to which people would go simply to not use Guix.
345.
by tucnak 1y ago
Surprised they never mentioned jina.ai models such as jina-embeddings-v3 at 8K context and outperforming most "contenders" in the MTEB or jina-clip-v2 (multimodal) or "late chunking," also known as mean pooling: https:&
346.
by tucnak 1y ago
Potato potato
347.
by tucnak 1y ago
You need LLM Ops. YC happens to have invested in Langfuse, which is if you're serious about tracking metrics, you'll appreciate the rest, too. And before you ask: yes, for cached content and batch completion discounts you can acco
348.
by tucnak 1y ago
Ask it to do less of it, problem solved, no? With tools like Cursor it's become really easy to fit the models to the shoe, or the shoe to the foot.
349.
by tucnak 1y ago
SWE's are learning about Type-2 tables the hard way round; of course, this is pure entertainment to data engineering people! Wait until "Audit table considered harmful" blog posts start popping up...
350.
by tucnak 1y ago
DELETE in SKIP LOCKED scenario is recipe for failure. God/Postgres gave you transactions for atomicity... only UPDATE w/ SKIP LOCKED, holy brother in Christ, and only DELETE completed normally.
351.
by tucnak 2y ago
Svelte now exists; it's a bit of a "clever" fullstack
352.
by tucnak 2y ago
There's foreign data wrappers for Clickhouse that still allow Postgres as single point of consumption with all the benefits of Clickhouse deployment. This is how we consume Langfuse traces!
353.
by tucnak 2y ago
The slides are a blast, thanks Classic mathematicians!
354.
by tucnak 2y ago
Everything is gambling
355.
by tucnak 2y ago
For sharing unique ideas
356.
by tucnak 2y ago
"If you're so clever, why are you so poor?" Ever heard this saying?
357.
by tucnak 2y ago
Common sense is a kind of skill. You don't have to make "optimal" trade, too, so the true dip doesn't matter only thing that matters is recognising that a ticker is undervalued. There's markets can be irrational ang
358.
by tucnak 2y ago
Why is that?
359.
by tucnak 2y ago
Honestly, HN is obsessed with finance; it seems at times like it's all you think about. But when it comes to basic trades, suddennly it's all excuses and "easy to time the market when you are from the future". A lot of p
360.
by tucnak 2y ago
When you say "buddies" what you really mean is "banks", "hedge funds" and "people to whom it was obvious that announced tariffs were bonkers". If you had the foresight and market sense, you could have

More ›