alach11

9 ms·

121.
by alach11 2y ago
For completeness, I tested o1-preview and o1-mini. Both were able to identify the language, but only o1-preview realized it was a sudoku solver.
122.
by alach11 2y ago
This is similar to Artifacts [0] in Claude. [0] https://support.anthropic.com/en/articles/9487310-what-are-a...
123.
by alach11 2y ago
It's pretty amazing that they made prompt caching automatic. It's rare that a company gives a 50% discount without the customer explicitly requesting it! Of course... they might be retaining some margin, judging by their discount
124.
by alach11 2y ago
All it speaks to is that tokenization is weird and introduces artifacts to LLM performance. Counting letters is a trivial task when you're staring at words on a screen. It's much harder when you're perceiving vectors based on
125.
by alach11 2y ago
I was surprised to learn that this is different from the similar-looking Pineberry [0]. [0] https://en.wikipedia.org/wiki/Pineberry
126.
by alach11 2y ago
This is one of the least interesting questions to ask LLMs. I wish it wasn't so popular.
127.
by alach11 2y ago
The phrase "turn down" is the opposite of "turn up". To "turn down" would be to decrease the intensity of the party. And "turn down for what" means something like "don't stop the party for a
128.
by alach11 2y ago
We have an amazing global experiment where we can watch different forms of governance succeed and fail. One of democracy's greatest failings is the slow creep of bureaucracy. The elevation of the self over society inexorably leads to a
129.
by alach11 2y ago
> I'm not convinced OpenAI is using one model. Look at the thinking process (UI), which takes time, and then suddenly, you have the output streamed out at high speed. According to OpenAI, the model does it's thinking behind the
130.
by alach11 2y ago
In the August 2nd, 2024 update on the GoFundMe: "Because our universities have generously funded our defense up to this point, we have not (yet) had to use any of the GoFundMe money... we are reaching out to you to explain how we inten
131.
by alach11 2y ago
I wish there were more details about the strategy/techniques he used. This reminds me of reading up on Cannonball runs [0] (racing coast-to-coast as fast as possible). Though this is a lot more pro-social! [0] https://arstec
132.
by alach11 2y ago
Anyone in industry is already making more than the NSA can afford to pay.
133.
by alach11 2y ago
Tying it to donations brings its own set of issues, such as disallowing tax write-offs due to a received benefit.
134.
by alach11 2y ago
Do you have benchmarks demonstrating this? In my own personal/team benchmarks, I've seen 4o consistently outperform the original gpt-4.
135.
Google uses AI to reduce stop-and-go traffic on your route(blog.google)
98 points by alach11 2y ago|125 comments
136.
by alach11 2y ago
We’ve saved a few million dollars with it at my job.
137.
by alach11 2y ago
It's also very likely that 3.5-turbo is more expensive for them to run than gpt-4o-mini. Models are getting smaller and more efficient. They just keep 3.5-turbo around for legacy support.
138.
by alach11 2y ago
Props to the author for releasing the data. My instinct is also to immediately suspect data leakage. It's super easy for this to happen. For example the original dataset could contain multiple articles about the same event.
139.
by alach11 2y ago
Claude 3.5 Sonnet took a solid lead on our internal benchmarks over gpt-4-turbo for extraction tasks against large documents. It may not be great for every workflow, but it certainly hits a sweet spot for intelligence x cost on most of my w
140.
by alach11 2y ago
I don't believe OpenAI supports batch calls for inference... only for embeddings. If you're interested in cost optimization, however, you're likely better off using Claude 3.5 Sonnet (as a stand-in for gpt-4o) or Claude 3 Hai
141.
by alach11 2y ago
You don't think compensation is the bigger issue?
142.
by alach11 2y ago
I use LLMs extensively in my field to automate all sorts of tasks. Need to classify a million PDF documents for cheap? Write a prompt and submit a batch job. Need to read 30,000 drilling reports to automatically scan for hazards? Done in 60
143.
by alach11 2y ago
There's no doubt that LLMs massively expand the ability of agencies like the NSA to perform large-scale surveillance at a higher quality. I wonder if Anthropic (or other LLM providers) ever push back or restrict these kinds of use case
144.
by alach11 2y ago
This is a quote, not a recording. Assange disputes the accuracy of the quote.
145.
by alach11 2y ago
With handwriting? With mixed fonts? Tesseract requires heavy customization and extension to perform reasonably on these workloads. The off-the-shelf options from major cloud providers blow it out of the water.
146.
by alach11 2y ago
Right. But I still have a lot of questions. How does the model handle when something important overlaps multiple tiles in high-resolution mode? Am I better off doing the tiling myself with some overlap?
147.
by alach11 2y ago
I really hope we see improvements to the resolutions large multimodal models can handle. Right now this patchwork approach leads to lots of unwieldly workarounds in applications.
148.
by alach11 2y ago
Seems like an interesting thing for Apple to do. Collect locations for phones that are marked stolen and look for dense clusters. Then work with local authorities to identify the perps. Would be a fun project for a small team!
149.
by alach11 2y ago
This is a fantastic article, and I've already shared it with a few colleagues. I'm wondering if you prefer to provide few-shot examples in a single "message" or in a simulated back-and-forth conversation between the user
150.
by alach11 2y ago
I'm on the (admittedly expensive) EGO system of lawn tools, and I can mow, trim, and blow my entire yard (front and back) in a relatively normal lot (8000 sq ft) on one 10.0Ah battery. It recharges in 2 hours. So I think electric is a

More ›