ankit219

18 ms·

211.
by ankit219 2y ago
Can you go into details (as much as you are comfortable) on what happened with the single tenant system? I have seen a few things, but I find it hard to put a finger on what went wrong except the ROI wasnt there. Would love to understand yo
212.
by ankit219 2y ago
Sorry for being vague. I meant LoRA, but used Apple as an example because their demo showed the potential. At a conceptual level, you can finetune a base model to be good at a specific task - eg: summarization, proofreading, generation etc.
213.
by ankit219 2y ago
Great article. I am not going to name names, but over the last one year, whenever there is a concept that became popular in Gen AI, thousands of startups pivoted to doing that. Many come from software background where the expectation was th
214.
by ankit219 2y ago
If this goes through, I don't see how this survives a court challenge. This is not a competition issue, but a data protection issue. Seems like it's brought under competition so that it spares the EU publishers who follow the same
215.
Why EU turned Meta's "Pay or OK" into a competition issue(medium.com)
3 points by ankit219 2y ago|0 comments
216.
by ankit219 2y ago
That is the contentious part. These are not easy questions. People talk about data, but individual data is as useless as they come for many companies. Data is only useful in aggregate and that too after reaching a certain scale. Counter int
217.
by ankit219 2y ago
Current laws are insufficient to address this, but we will end up using them to adjudicate whether or not this is fair use. We will get there in due time, and the pathway will be messy. I am ambivalent about this overall, but few things are
218.
by ankit219 2y ago
Have they ever specified any reason why they prevent users from finding this out? I read somewhere (I think Epic v Apple) where the judge surmised that App store model was not just payments but the whole mechanism of tracking subscriptions,
219.
by ankit219 2y ago
This is remarkable. If things like these could be AI generated (or even tweaked) to fit a use case, that kind of spells doom for bloated Saas products. Would be amazing to see the progress along this vector. You may want to look at https:&
220.
by ankit219 2y ago
In my limited (to one person) experience, it may not be as clear cut. I am one of those who uses language to think. I would have an internal monologue and I would also say things out loud just to see how they sound. It's not always a f
221.
by ankit219 2y ago
The problem is not just about interoperability. According to DSA, given Apple is one of the designated companies, they are responsible for security and privacy while DMA mandates that they allow for interoperability when they add functional
222.
by ankit219 2y ago
MMLU is not a reasoning benchmark. It's a measure of how distributed and representative their training data was and how well it's able to recall (for lack of a better word) based on training epochs. GPQA etc. test reasoning in som
223.
by ankit219 2y ago
This is likely more serious and substantiated (to an untrained eye) than I expected from just reading the tweet. Coverage on Github: https://github.com/OpenBMB/MiniCPM-V/issues/196
224.
by ankit219 2y ago
Everybody loves some sort of drama. If they don't find it, they invent it.
225.
by ankit219 2y ago
I understand where you are coming from, but I kind of see where the misconception comes from too. Over the last year and a half, "training" has come to mean everything from pretraining an LLM from scratch, instruction tuning, fine
226.
by ankit219 2y ago
Will share a longer post on this once I finish it. I have tested this multiple times, on bigger models, on custom smaller models, and it does not work. In a strict sense, finetuning can add new knowledge but for that you need millions of
227.
by ankit219 2y ago
Finetuning would never add any new knowledge. For internal corporate data, use RAG or train a model from scratch. Finetuning would not help anyone answer a question from that data.
228.
by ankit219 2y ago
I like the web feature, specifically when you are researching. Can understand the "hiding" at this point, as I guess they are still figuring out the best place for it. Sundar Pichai in some interview said that generative search on
229.
Cosmic-mapping project seems to show that dark energy could be weakening(nature.com)
2 points by ankit219 2y ago|0 comments
230.
by ankit219 2y ago
Your experience is very helpful. I took a theoretical example since where I live breath analysers are how you get fined for drunk driving. I didn't think it was possible to contest that. I would rather that these things happen preempti
231.
by ankit219 2y ago
I am not comparing it to a human error. There is a conventional wisdom that humans make mistakes, and we have redundancy built into various systems to counter the same. With machines and tech, the expectation is that they are accurate (whic
232.
by ankit219 2y ago
When i read these implementations, one thing I think about is the British post office scandal, also known as Horizon IT scandal[1]. These technologies are probabilistic with some nonzero chances of error which can't always be easily de
233.
by ankit219 2y ago
These bigger companies are releasing open source models for publicity. For Databricks and Snowflake, both want enterprise customers, and want to show they can handle swathes of data and orchestration jobs, what better way to show that than
234.
by ankit219 2y ago
Not trying to disparage them, but their models always give a feeling that it is overfitted on benchmarks hence they perform so well. On everyday tasks, it's much worse - chat or simple completion tasks. Distilling can work and there ar
235.
by ankit219 2y ago
I am not coming from a jaded bootstrapper angle at all. My point is simple: if you are in fundraising mode and are fundable, apply to YC. And talk to as many investors as you can. If you are not, don't do any of those. Focus on buildin
236.
by ankit219 2y ago
Fundraising is about focus. If you are in a fundraising mode and applying to YC, great. Would learn a lot, and I think it makes sense to do that. When fundraising, many rejections are great. When not fundraising and a VC reaches out, it&#x2
237.
by ankit219 2y ago
One of PG's blogs talk about being fundable. For a founder, there is always a transition from being a builder to being an entrepreneur (business owner) where you must be able to access whether you are fundable and know what you need to
238.
by ankit219 2y ago
I find these kind of messages distasteful. When people are focused on building, they need feedback from the market. That rejection/feedback helps, but any other kind of rejection - especially a VC or investor rejection - hurts. You cou
239.
by ankit219 2y ago
Seems like a false equivalence and trying to forcefit two things. It's not a binary choice as the writer puts it. Apple's tracking transparency gave users a choice whether to enable tracking/analytics outside the app usage. (
240.
by ankit219 3y ago
I can understand ATS is dehumanizing, but to frame it like in the article does little to solve the problem. The article does not even highlight the actual problem as to why companies have such automated processes. Number of applicants per j

More ›