zurfer

18 ms·

241.
by zurfer 3y ago
I would love to bring your attention also to getdot.ai We launched it on Hackernews with an analysis of HN post data. https://news.ycombinator.com/item?id=38709172 the main problems we see in the space: 1) good interface de
242.
by zurfer 3y ago
as another sibling points out, this seems too high. Their official website speaks about millions of lines of code [1]. They also published an article (2017) where they proudly say, they got rid of half a million lines of code [2]. A comment
243.
by zurfer 3y ago
Was it a weekend project? ;) looks cool!
244.
by zurfer 3y ago
The potential application that I'm most excited about is better high bandwidth networking. In data centers e.g. for training on multiple GPUs this could yield a magnitude improvement in shuffling data around. If I read it correctly the
245.
by zurfer 3y ago
Great question! Here is Dot's answer: https://eu.getdot.ai/share/4e80130e-9c0c-4a62-b382-402a978e0... We actually wanted to post it as Show HN, but dang advised that since it is not directly interactive it is more
246.
by zurfer 3y ago
Python is hard, right now we utilize OpenAI's code interpreter for that, which isolates the workloads pretty well at the cost of speed. I'm hoping we can improve that trade-off in the coming weeks.
247.
by zurfer 3y ago
In general that sounds more like a classical search engine to me. Here I would check out Algolia ( https://www.algolia.com )
248.
by zurfer 3y ago
Thank you for sharing that. Building the training space for Dot is definitely an important part of the value prop.
249.
by zurfer 3y ago
It is interesting to see that exponential growth stopped and linear growth set in. I wasn't around at that time, so I'd be curious if it's the reason you give. For me HN is the highest signal place on the internet for founder
250.
by zurfer 3y ago
That's along the lines of what Dot is doing. For the full details you can expand "Explanation > Full Logs".
251.
by zurfer 3y ago
That's exactly what Dot is about. Preprocessing goes into data transformation territory and given the millions of comments on HN it's also getting expensive quite fast. From BigQuery and Snowflake I know they have remote/exte
252.
by zurfer 3y ago
We will probably enable the Hackernews data as demo data in the future. Today we only have some e commerce demo data when you sign up. Although you could connect the public Big query dataset and do it yourself. Your 2nd question would requ
253.
by zurfer 3y ago
The biggest project I am aware of is the SQL agent from langchain. It definitely gets you started and is great imo for single developers or small technical teams.
254.
by zurfer 3y ago
Great question. We allow Dot to work with just meta data, but you can enable it to also react to the content and we recommend to our customers to also pass content to the LLM (Azure GPT4) because it's a lot more capable, e.g. for filt
255.
by zurfer 3y ago
Thank you for the feedback! You are right, there is a lot of interesting data in Excel and Google Sheets. Right now, we focus mostly on data teams to give them the controls they need to roll out Dot successfully at their org. But yeah, we c
256.
by zurfer 3y ago
Security is an interesting challenge. The way we approach it is that we assume the LLM will spit out actions that are wrong or harmful. So everything needs to be handled with old school permissions. Dot has a technical user that right now
257.
by zurfer 3y ago
Thanks! You probably refer to one of the demos on our landing page? I like how you describe the problem. You're absolutely right that SQL seems easy but it's these edge cases that make it hard to get right. Joining metrics with a
258.
by zurfer 3y ago
In our experience Dot can come up on the fly with cohort analysis charts if the underlying data is well structured. In most cases however, some level of explanation, example or data preparation is needed for robust and repeated cohort analy
259.
by zurfer 3y ago
https://eu.getdot.ai/share/f94644f5-c76a-4211-9c48-76800f12f... Not really :) Dot will ask a clarifying question. Questions like, "what kind of data do you have access to?" or "how can you help me?"
260.
by zurfer 3y ago
Thank you! Large models and analytics have a great match and will amplify how we can work with data.
261.
Hacker News Activity Analysis with GPT-4 Agent(eu.getdot.ai)
139 points by zurfer 3y ago|51 comments
262.
by zurfer 3y ago
People tell you that something did not work, after you ask them? I need to manage an email sequence with 8 follow ups to get explanations.
263.
by zurfer 3y ago
So 14400 H100 for GPT-4, but that's just a fraction of the new system that Azure is building for OpenAI. FWIW, I most enjoyed the 29TB machine demo at the end.
264.
by zurfer 3y ago
Thank you for linking that. It is worth reading. I'm always amazed how cheap and well China can manufacture a wide variety of devices. There is ofc a lot of expertise and economies of scale at work, but the letter is a sad reminder tha
265.
by zurfer 3y ago
This seems to refer to the GSM8K (grade school math) benchmark [1]. GPT-4 scores 0.92 on that one. A breakthrough could mean it gets all of them correct. This would have major implications for long-term planning. For instance, if you have a
266.
by zurfer 3y ago
I don't buy it. If you listen to Sam and Greg, they want to change the world and bring us into a world of abundance that everyone can participate in. They were the ideators of this "awkward non-profit owning a for-profit company&q
267.
by zurfer 3y ago
as others pointed out, it's important how you ask and what you think the truth is. https://chat.openai.com/share/cc688592-236a-499b-82e1-81bdaa...
268.
by zurfer 3y ago
But is that really the case? It's actually not just a database lookup. It is abstracting and compressing throughout the training process. And it clearly learns certain rules about our world. It seems LLMs can reason, but reasoning is a
269.
Ask HN: Why is GPT-4 or Claude-2 so bad at tic-tac-toe?
2 points by zurfer 3y ago|7 comments
270.
by zurfer 3y ago
Benchmark it with the standard YC deal. 125k USD for 7%.[1] Now YC adds value on top (unlike your typical angel), but your average startup also doesn't get into YC. The main questions is: Can the team you're investing in build a b

More ›