ankit219

14 ms·

151.
by ankit219 1y ago
It's 100% Google saves my data. And gives me an option to delete that data. It uses that data, and while I have some degree of control over it, i don't have full control over it. As a user it's preposterous that any kind of d
152.
by ankit219 1y ago
I might be missing something big here, but how does the consistency work in these scenarios? At one place, you are saying Google is not allowed to pay to be default, but at the same time you are allowing others to be default because they ar
153.
by ankit219 2y ago
Agents in the current format are unlikely to go beyond a current levels of reliability. I believe agents are a good use case in a low trust environments (outside of coding where you could see the errors quickly with testing or deployment) l
154.
by ankit219 2y ago
You don't have to tolerate agent/AI going haywire. In a simple example, say of multiple parallel generations. It's compute intensive and it reduces the probability of your agent going haywire. You need mechanisms and evals t
155.
by ankit219 2y ago
Thats the core of the argument. We are switching from a 100% deterministic and controlled worldview (in software terms) to a scenario where it's probabilistic, and we haven't updated ourselves accordingly. Best of n (with parallel
156.
Bitter Lesson is about AI agents(ankitmaloo.com)
139 points by ankit219 2y ago|105 comments
157.
by ankit219 2y ago
You can take an obviously obtuse and extreme interpretation to what I said. What i mean is simply this: Today, there is an implicit social contract that if i build a better tech or innovate in a niche, i get to monetize it anyway i like. In
158.
by ankit219 2y ago
This isn't true. They complied with most of it. This is a guidance in EU-DMA parlance, where EU is specifying[1] how it wants Apple to comply, not that apple hasnt complied and getting punished. Those woudl have been fines. From the do
159.
by ankit219 2y ago
Wouldn't other companies run into similar issues given they know whatever they make could be used by other companies for free if they become big enough?
160.
by ankit219 2y ago
I have seen this analogy before, and (hence the question). Apologies if it's rude. By my understanding, while the tools are important, most of the apps hit escape velocity as the underlying models became good enough. You had cursor doi
161.
by ankit219 2y ago
We did a research in this regard: https://arxiv.org/abs/2409.17171 The idea is to create bespoke models for org at 90% lower compute. (we cheat a little, where we use an underlying open source model and freeze the exis
162.
by ankit219 2y ago
answering the first question if i understand it correctly. The missing piece is data obviously. With search and code, it's easier to get the data so you get such specialized products. What is likely to happen is: 1/ Many large com
163.
by ankit219 2y ago
I did not mention apple in my comment. And what you wrote does not contradict what i wrote. If they truly started with customers, perhaps we can agree that having to click another site to get to a simple one line answer is wrong. Thats why
164.
by ankit219 2y ago
So, after the first round of compliances, they did a review of the changes and did meetings with the likes of Yelp etc. This led to the next round of changes.[1] I am for the changes, just that I think you should also look at it from user p
165.
by ankit219 2y ago
Supporting a policy or a regulation which looks at just competition with zero incentives to care for users because it may help the portfolio is not a good look. People may argue otherwise, but DMA exists to make sure the gatekeepers lose ma
166.
by ankit219 2y ago
Think this captures one of the bigger differences between what Open AI offers and what others offer using the same name. Funnily enough, Google's Gemini 2.0 Flash also has a native integration to google search[1]. They have not done it
167.
by ankit219 2y ago
Not really. It could be that "I would orient my marketing efforts to get you 1000 leads in 2 months. If I hit the target, i get x amount, else y based on the final result. You dont need the percentages. The key is to position yourself
168.
by ankit219 2y ago
You can control this. You could opt for the part where the dependence on rest of the team is minimal. Or you could structure it in a tiered manner. With different pricing at different levels of outcomes.
169.
by ankit219 2y ago
One thing i have seen while freelancing (on mostly marketing tasks) is how people are willing to pay a lot if you own the outcomes. The typical model is work per hour, but if you can own some outcomes you control aka drive value, many busin
170.
by ankit219 2y ago
I said so too, I used google instead of gemini. Somehow it did not create as much of a buzz then as it did now.
171.
by ankit219 2y ago
I think it's bound to underwhelm the experts. What this does is go through a number of public search results (i think its google search for now, coudl be internal corpus). And hence skips all the paywalled and proprietary data that is
172.
by ankit219 2y ago
Every time OpenAI comes up with a new product, and a new interaction mechanism / UX and low and behold, others copy the same, sometimes leveraging the same name as well. Happened with ChatGPT - a chat oriented way to use Gen AI models
173.
by ankit219 2y ago
literally this: https://support.google.com/websearch/answer/13572151?hl=en&c...
174.
by ankit219 2y ago
At this point, it's a function of how many thinking tokens can a model generate. (when it comes to o1 and r1). o3 is likely going to be superior because they used the training data generated from o1 (amongst other things). o1-pro has a
175.
by ankit219 2y ago
Yes that is true. Though that can be subsumed if you notice it, and ask the model to ignore those biases. (an extreme example would be opposition prep for a debate). I am not interested in politics and other related issues anyway.
176.
by ankit219 2y ago
There are two aspects to this from my pov. And I think it might be controversial. When i have a question about any topic, and I ask Chatgpt, i usually chat about more things, coming up with questions based on the answer, and mostly stupid q
177.
by ankit219 2y ago
> The other thing was that o1 had access to many more answer / search strategies. For example, if you asked o1 to summarize a long email, it would just summarize the email. QwQ reasoned about why I asked it to summarize the email. O
178.
by ankit219 2y ago
There is remarkable paper pro, but it's expensive. And i would want a browser at the very least to download stuff for my reading. (mostly pdfs)
179.
by ankit219 2y ago
No worries. If you need any specific help with the use case, please feel free to reach out on my email. ankit at clioapp.ai
180.
by ankit219 2y ago
Reading this, thinking in terms of moats is useful, but in terms of AI, we are not there yet. There is a promise of exceptional improvement to everything, so much that many companies which takes ages to change a software are moving at a sig

More ›