leobg

24 ms·

331.
by leobg 2y ago
I’m willing to bet that I can reduce your costs by at least 10x. I’d go so far as to say this thing should be able to handle HN front page traffic at < $300 / month, including all real-time vector search. That is, if this 6k number
332.
by leobg 2y ago
The weird thing, to me, is that it should be much more fulfilling to do one thing well than to pretend to be doing ten things and doing them badly. The former is the attitude of the craftsman. The latter the attitude of a slave.
333.
by leobg 2y ago
Could you not go a step further and use gpt4o-mini’s vision to check if the text is readable? Basically, extract the text from the biggest screenshot. Then compare it to extracts from increasingly smaller versions until the text degrades &#
334.
by leobg 2y ago
It’s not just the U.S.. Here in Germany, there are several multi million Euro fraud cases that have been going on for decades. They get picked up by the media every couple of years, with hidden camera footage and everything. But the authori
335.
by leobg 2y ago
If that shouldn’t be called “technical debt” I don’t know what else.
336.
by leobg 2y ago
That blog is incredible. Thanks for sharing.
337.
by leobg 2y ago
Same here on anything Elon. HN is like an uncle who knows a lot and teaches you new things every time you hang out with him… but who also has a few really weird sore spots that you better never mention in his presence.
338.
by leobg 2y ago
Well, it really doesn’t suit itself for automation, much less for real-time transactions. Say, you want to sell access to an article on your website. It would be cumbersome for the customer to have to switch over to their online banking and
339.
by leobg 2y ago
The problem is that it’s almost impossible to teach knowledge to an LLM without teaching a specific form of expression at the same time. When you have a question/answer tuple in your training data, you are also teaching the model that
340.
by leobg 2y ago
Article is an excerpt from this book: Theodore H. Schwartz. “Gray Matters: A Biography of Brain Surgery” (2024)
341.
by leobg 2y ago
What about Google itself? Store content from billions of websites in your own index without permission and show snippets from those sites to your own users, making yourself into “the source” of all answers when, in reality, it’s all just sc
342.
by leobg 2y ago
In the Steve Jobs biography, the author notes that Steve believed his cancer to have been caused by a particularly stressful time in his life, even though a scientific link between stress and cancer had never been found. I’m quoting from me
343.
by leobg 2y ago
Well, it was before he met his new wife, wasn’t it? It seems to me that coincided with his shift of priorities.
344.
by leobg 2y ago
Just imagine how different the Internet would look like today if receiving payments would’ve been as easy as receiving email from the beginning. That it is not trivial for a single person on the Internet to receive payments without a third-
345.
by leobg 2y ago
> Entropy-weighted similarity: We adjust the similarity scores between query tokens and related documents based on the entropy of each token. Sounds a lot like BM25 weighted word embeddings (e.g. fastText).
346.
by leobg 2y ago
Just had this exact discussion with the kids at dinner yesterday. In our case, we talked about the simplified traffic light rule “go on green, stop on red“. This to come up is wrong, of course. Because if the light turns red while you are c
347.
by leobg 2y ago
A cheap DIY way of achieving the same thing as RLHF is to fine tune the model to append a score to its output every time. Remember: The reason we need RLHF at all is that we cannot write a loss function for what makes a good answer. There a
348.
by leobg 2y ago
Joe Cocker spent the final years of his life growing heirloom tomatoes. https://www.denverpost.com/2015/02/24/joe-cockers-crawford-e...
349.
by leobg 2y ago
Actual content: > Calming activities reduced anger in the lab and the field, they found, and across other variables like methods of instruction or participant demographics. Effective arousal-reducing activities included slow-flow yoga, m
350.
by leobg 2y ago
That would mean the winners will be those who turn towards a local community. Which makes sense. I often wondered how lawyers who are terrible at their craft can stay in business. Thing is, they just don’t compete. They build relationships
351.
by leobg 2y ago
I think that was beautifully phrased. Needless to say, I agree.
352.
by leobg 2y ago
+1. What city?
353.
by leobg 2y ago
Why is it called Black Forest Labs? Are you in the Schwarzwald? Your names sound German, but I see no imprint.
354.
by leobg 2y ago
The relevant section from the article: > [C]onsuming foods and beverages sweetened with xylitol can make the platelets in your blood more likely to clot — a recipe for serious heart events. > In studies performed by Dr Hazen’s team, h
355.
by leobg 2y ago
Changes nothing. Thief: I’m not guilty. There is no free will, so I couldn’t have acted otherwise. Judge: I sentence you anyway. Because my actions, too, are predetermined.
356.
by leobg 2y ago
I wanted to try it. But sharing my phone number before I even know what the app does? No way. I also don’t see why you would need it. I think “Riffusion” is a cool name.
357.
by leobg 2y ago
Sharp as a tack.
358.
by leobg 2y ago
So why don’t we have Sketchpad on our iPads today?
359.
by leobg 2y ago
Voice Dream Reader. Using it for 10+ years. Haven’t found anything that comes close.
360.
by leobg 2y ago
Anything like that available for iOS?

More ›