zurfer

17 ms·

211.
by zurfer 2y ago
As someone who has build a data catalog tool and is now building one of these looked down on LLM agents(1): Data catalogs are mostly used by technical people/the data team and a handful of power users. Why? Because it's intellectu
212.
by zurfer 2y ago
502 :( I wonder if it's something similar to what I did some months ago: https://eu.getdot.ai/share/c80139c9-13f4-4db4-88f6-6e058ba31...
213.
by zurfer 2y ago
"Four feet is equal to four feet. The measurement does not change as it is already in the desired unit." gpt4o
214.
by zurfer 2y ago
looks great. Although in your example it seems like the y-axis is a bit off. start: 0.05 vs 0.1 end: 0.65 vs 0.7 It's a difficult example!
215.
by zurfer 2y ago
I share the authors view that if things align and compound we are in for a very wild ride. However, I feel the bull case is over emphasized. Not all likely "walls" are sufficiently discussed, some that I think about: - will AI blo
216.
by zurfer 2y ago
the other comments get it. It means that data comes in a ton of different shapes with poorly described schemas (technically and semantically). From the typical CSV export out of an ERP system to a proprietary message format from your own cu
217.
by zurfer 2y ago
This is not fully correct. Originally big data was defined by 3 dimensions: - Volume (mostly what the author talks about) [solved] - Velocity, how fast data is processed etc [solved, but expensive] - Variety [not solved] Big Data today is n
218.
by zurfer 2y ago
That's one of the more feature rich AI analytics assistants. (1) Kudos for open sourcing. I think it's really difficult to build a business around that, but there are some successful examples in the space: metabase, airbyte, dbt,
219.
by zurfer 2y ago
This is cool and in my mind super useful for migrations. It seems the main benefit of using something like that in daily life is that it's more convenient to generate complex SQL statements (like pivoting a table with a lot of columns)
220.
by zurfer 2y ago
it's a good article, but misses the point. you can't apply the scientific method early in a startup. You never have enough data and all the factors included change too quickly. what you need is conviction. if it works for 1/1
221.
by zurfer 2y ago
Congrats on shipping something and getting into the app store, but I would never use it. Reading to my kid is super precious and should not be automated. I rather have something that helps me with spending time with him (it's probably
222.
by zurfer 2y ago
It's fair to point that out, but the "plan" right now is to build AGI that is able to do a ton of things. Current frontier models are not optimized to be best in class translators, they just happen to be really good at that
223.
GPT-4-turbo-2024-04-09 "wins" simple evals benchmark(github.com)
2 points by zurfer 3y ago|0 comments
224.
A list of AI analytics tools(github.com)
4 points by zurfer 3y ago|0 comments
225.
by zurfer 3y ago
thank you, I was really confused by the chart. It's kind of odd that you would use 1 as a nominator. both indexes next to each other are maybe more intuitive: https://fred.stlouisfed.org/graph/?g=1jzxK
226.
by zurfer 3y ago
Every big tech in the beginning looked fragile/no moats. I think we'll see that data, knowledge and intelligence compound and at some point it will be as hard to penetrate as Meta's network effects.
227.
by zurfer 3y ago
Barrier to entry with "few $B" is pretty high. Especially since the scaling laws indicate that it's only getting more expensive. And even if you manage to raise $Bs, you still need to be clever on how to deploy it (talent, c
228.
by zurfer 3y ago
I'm really surprised by the variety. Having lived in Berlin for 6 months I completely missed that detail.
229.
by zurfer 3y ago
Data is for seeing problems, not for finding solutions. As by the authors example, you lose 10M in fraud? Good that you monitor that! Otherwise it would be hard to justify spending time on it.
230.
by zurfer 3y ago
I also love how Observable Plot looks, but agree with top poster, the things that keep me from switching are: - Python wrapper - Out of the box interactivity
231.
by zurfer 3y ago
We are building Dot ( https://www.getdot.ai/ ). We mostly focus on good conversation design and data governance to ensure a great end user experience. Here is an example conversation about HN data https://eu.getdot
232.
by zurfer 3y ago
really cool, the license is not really standard, but seems open source. The actual model can be found here: https://huggingface.co/cfahlgren1/NaturalSQL-6.7B-v0 This seems like a great base model, although I wonder if
233.
by zurfer 3y ago
Is the Salesforce tower not considered to be a part of Silicon Valley?
234.
by zurfer 3y ago
Thank you! I'm not experienced with 7B models. 3 things stand out to me: - it's absolutely not useable for the kind of use cases I solve with GPT-4 (code generation, information retrieval) - it could technically swallow a 50 page
235.
Gemini Benchmark – MMLU (compared with GPT-4-turbo, Mixtral)(hub.zenoml.com)
1 points by zurfer 3y ago|1 comments
236.
by zurfer 3y ago
Most interesting is how they perform by subtask. E.g. gpt-4-turbo vs gpt-4-turbo with chain of thought (cot) goes from 0.51 to 0.88, but sometimes cot makes things worse (e.g. in management).
237.
by zurfer 3y ago
For me the main question is not, is the code good enough or will competitors copy, but how can we make enough money to build a sustainable business? The quoted and beloved open source projects are not good businesses: PostgreSQL, Python, Bi
238.
by zurfer 3y ago
This is great. Thank you! I would be especially interested in more details around speed. Average is a good starting point, but I would love to also see standard distribution or 90, 99 percentiles. In my experience speed varies a lot and it
239.
by zurfer 3y ago
as someone building in this space, I am a bit surprised how many concepts you managed to combine in your last sentence. :'D I will bookmark: ... and we will move our own built-in agents to it as well (ex: GPU-accelerated dashboard pane
240.
by zurfer 3y ago
that is correct. GPT-4 is good on well-modelled data out of the box, but struggles with a messy and incomplete data model. Documenting data definitely helps to close that gap. However the last part you describe is nothing new (BI teams taki

More ›