maytc

All the apps on the app store wanted to charge me $20/mo to upload a pdf and export as SVG. So... I built a free one.

Most code written are not novel. Actually most code written should not be novel. Eg: the number of lines of code written isn't spent on writing git, it's spent on writing that landing page no one ever sees.

The key difference is that the model is able to write the program as it’s executing it.

Before it needs to write the code and have an external program execute it. Here it can change its mind mid execution. Kinda like what was observed in the CoT’s ah ha moment

Was hoping to upgrade from my 2017 iMac and this new studio display is quite a bummer. 9 years later and it's basically the same display spec. (500 nits spec bump to 600 nits)

It’s probably because their paid advertising spots are being viewed by bots instead of humans. Amazon still charges the advertisers, since it detects the traffic as coming from a non-bot user based on the user agent.

They fixed it in the press release charts https://openai.com/index/introducing-gpt-5/

2400+ fps (mindblown)

Curious if the authors had a chance to look at the Softpick paper? https://arxiv.org/abs/2504.20966

That example is probably in the training data?

The puzzle assumes that the room temperature is greater than the cold milk's temperature. When I added that the room temperature is, say, -10 °C, Mercury fails to see the difference.

The underrated release is the ChatGPT search Chrome extension that lets you bypass Google search completely.

https://chromewebstore.google.com/detail/chatgpt-search/ejcf...

Switched to Perplexity Pro and haven't looked back

The difference in the dates example seems right to me 20 October 2024 and 2024-20-10 are not the same.

Months in different locales can be written as yyyy-MM-dd. It can also be a catalog/reference number. So, it seems right that their embedding similarity is not perfectly aligned.

So, it's not a tokenizer problem. The text meant different things according to the LLM.

A tool to auto-translate images and PDFs.

Curious if anyone got reasonably good at a foreign language through duolingo? (Maybe able to hold a conversation)

Github Copilot definitely improved my coding productivity. However, most of my time is spent thinking about the problem to be solved than the actual coding.

Link to the paper: https://arxiv.org/abs/2311.18828

IMHO, cascading is OKRs' real power. It forces the organization to say no to most things, which sharpens its focus on what needs to be done.

Throw that away, setting aligned SMART goals would achieve a similar effect as aligned OKRs.

Seems like a typo. Other pages says GPT-4 Turbo https://www.microsoft.com/en-us/store/b/copilotpro

Stupid question but I thought transformers have an O(n^2) memory usage. With 2M tokens, won't I need dozens and dozens of GPUs just to run the base LLaMA2 models?

Helping teams know what to say "no" to is the real power of OKR.

You do that by having the entire organization's O and KR roll up and cascade down. My Objectives directly roll up to my parent organization's Key Result. My parent org's Objectives rolls up to their parent's KR, and so on to the top. Then, you have the top-level check downwards if the sets of OKRs still make sense in their entirety.

Unfortunately, I have never seen any organization I was with ever do this...

Built an LLM interface to control my browser. Used it to generate playwright tests for me https://github.com/mayt/BrowserGPT

Never liked the way these problems are worded.

`You take a random ball out of the urn—it’s red—and discard it.`

How normal people read it: Given this specific instance where you just discarded a red ball from this urn, what's the probability of the next ball?

How it expects you to read it: Given infinitely many random samples from the urn. For cases where you get red, remove it, then take a second sample. What's the probability of the next ball, given all the samplings?

Can you add a knob to increase or decrease the blur?