billythemaniam

7 ms·

31.
by billythemaniam 4y ago
Specifically on high/low income intertwined, yes. There is certainly some segmentation, but there are also places where one street is good and the very next street is bad and the next street is good. It's an interesting city for s
32.
by billythemaniam 4y ago
I don't. I am very happy to not use Vitess and more than willing to recommend your competitors. Don't feel bad.
33.
by billythemaniam 4y ago
I didn't claim that I am dealing with a gigantic workload (total QPS is a lot higher btw). Rather I claimed we are at the point where a distributed database is a good idea if the business wants to continue to scale without major databa
34.
by billythemaniam 4y ago
I'm not sure what "more proven" exactly means. If you mean, it is not as well known and used in US? Yes that's true because the core developers are Chinese. If you mean, it hasn't been used at the same scale in prod
35.
by billythemaniam 4y ago
Spanner and Cassandra really shouldn't be in the same sentence. They are optimized for very different use cases. The "obsolete" part of that quote does apply to Spanner, TiDB, and CockroachDB in my experience. I haven't
36.
by billythemaniam 4y ago
At least in Tallinn, most people speak some English. I met a surprising number of Estonians that were fluent too. There seemed to be a love affair with New York City, I saw New York pizza and clothes advertised everywhere in Tallinn.
37.
by billythemaniam 4y ago
If ES doesn't work for you, I recommend Vespa. https://github.com/vespa-engine/vespa Others have made other suggestions, but Vespa has two unique features. First it is battle tested at a large scale, second it sup
38.
by billythemaniam 4y ago
ES has support for vector search now too. Really you want both in use cases where the user expects the the top results to contain the search keywords, but also wants results that are synonyms or conceptually similar. TF/IDF and BM25 he
39.
by billythemaniam 4y ago
So far GPT-4 seems to improve on this problem. Still happens but less frequent.
40.
by billythemaniam 4y ago
Yeah you can copy/paste some text then ask it to summarize it, extract something from it, ask questions about it, and probably other things.
41.
by billythemaniam 4y ago
Definitely true for GPT-3, but after using GPT-4, I am not so sure. I am sure there are cases where a specialized NLP model can beat GPT-4, but I think they might be in the minority now. Whether I tried summarization, data extraction, or co
42.
by billythemaniam 4y ago
If you had full source code that fit into the context, do you think it could reliably answer questions about the code, build unit tests, generate documentation? I ask because that is the software equivalent of what you just described.
43.
by billythemaniam 4y ago
CS program, but depends on the classes you take. For me, the main class was called "computer architecture". Classes about compilers, databases, servers, embedded programming, server clustering can all help to understand the machin
44.
by billythemaniam 4y ago
It's certainly not quite as good out of the box, at least the open sourced checkpoints. However so far I found it can achieve similar accuracy with enough examples and/or fine-tuning for my use cases. Like everything, it depends o
45.
by billythemaniam 4y ago
Flan-T5 is much smaller than GPT-3, but was trained on significantly more data resulting in competitive accuracy. It is also Apache licensed. I wonder if that model is fast enough for enough use cases to make it cost effective?
46.
by billythemaniam 4y ago
I don't think it is necessary to understand how the internet works technically to "understand the internet". It's like saying a race car driver can't drive fast without understanding the physics of a piston.
47.
by billythemaniam 4y ago
Neither does yours, neither does mine.
48.
by billythemaniam 4y ago
I agree that the state of the art isn't ready yet for general consumption. I think GPT-3, etc are good enough to help with a wide range of tasks with guardrails. To clear, I don't mean guardrails around racist language, etc which
49.
by billythemaniam 4y ago
Let's talk about what ChatGPT (or fine-tuned GPT-3) actually is and what it is not. It is a zero-shot or few-shot model that is pretty good at a variety of NLP tasks. Playing tic tac toe or chess is not a traditional NLP task so should
50.
by billythemaniam 4y ago
Define "fast" please.
51.
by billythemaniam 4y ago
In terms of speed: int > float > decimal. Depends on the hardware type. On GPUs, float > int. However the performance difference is negligible for many, many use cases so I generally use decimal as the default and only use float if
52.
by billythemaniam 4y ago
What was their argument? I think it depends on how the answers from davinci are used. If the answers are used verbatim to fine-tune a cheaper model, then that could be problematic. However if it is used to generate a fine-tuning dataset whi
53.
by billythemaniam 4y ago
I don't really see the issue. You are using a service called GhostWrite which uses ChatGPT under the hood. OpenAI/ChatGPT would be considered a sub-processor of GhostWrite. What am I missing?
54.
by billythemaniam 4y ago
As someone who has written 100s (maybe 1000s at this point) of DBT models, the amount of Jinja you need is at most 1-2% of the codebase.
55.
by billythemaniam 4y ago
> One example is dbt. They started as writing simple SQL models. With the introduction of Jinja, majority of models look nothing like SQL anymore. You need to visualize your model to understand relationships. It took the beauty of SQL an
56.
by billythemaniam 4y ago
If you want to do it all in SQL, DBT gives similar capabilities. Modern data warehouses + DBT is really why there is a lot more "just use SQL" talk in the data engineering world recently.
57.
by billythemaniam 4y ago
While it matters, +-100ms doesn't matter in the vast majority of web use cases. There is even such a thing as too fast. If a user doesn't notice the page updated due to speed, that is also a poor experience. 300ms would actually f
58.
by billythemaniam 4y ago
I was trying to get davinci-003 to convert text to SQL, and it worked with a very simple prompt like "convert this text into SQL". With all their other models, I could get it to work too but all required a few examples within the
59.
by billythemaniam 4y ago
DBT has two mechanisms to implement the blog post. The first is macros which closely resembles m4. The second is "ephemeral" models which are separate SQL files which get compiled into a CTE when referenced in the main model.
60.
by billythemaniam 4y ago
For me, I would much rather SQL especially when using DBT too to manage the queries like any other language. I thought the same way as you and SQL certainly has its warts, but I have grown to appreciate it's elegance. I think over the

More ›