ericlewis

no.

much simpler, just pull apart the MT VST :P

I am working on PO-33, doing EMFI.

I still dont understand why people point to this chart as any sort of meaning. Cost per task is a fairly arbitrary X axis and in no way representing any sort of time scale.. I would love to be told how they didn't underprice their model and give it an arbitrary amount of time to work.

US, GA here. My mom was big on tanning and warned of us this (lemons are also bad). I believe she said something about it being used on purpose for tanning, but that you had to be careful or you would badly burn. She probably did that around the late 80s or early 90s.

As a person who works on this kinda stuff I know what they mean. It’s very hard to debug things totally blind.

Higher the precision the better. Use what works within your memory constraints.

If you are using the user attribution with OpenAI (as you should) then they will block that users id and the rest of your app will be fine.

The theory is that you increase the context with more relevant tokens to the problem at hand, as well as its solutions, which in theory makes it more likely to predict the correct solution.

This reminds of a TPU.

I have one of these. It’s not there.

Perhaps it’s because I know human beings that have the exact same operation and failure mode as the LLM here and I’m probably not the only one. Failing at something you’ve never seen and faking through it is a very human endeavor.

My wife is capable of “automatic driving” and I am not. She describes it as a flow state.

You would need both. RAG will have to be part of the process if you want to generate or work on an existing project. Since it stands for Retrieval Augmented Generation. A fancy way of saying an LLM can access documents.

GPT-4 is fine-tunable.

https://platform.openai.com/docs/guides/fine-tuning

“ Fine-tuning for GPT-4 is in an experimental access program - eligible users can request access in the fine-tuning UI when creating a new fine-tuning job.”

You need only have fine-tuned a 3.5 model ime.

ChatGPT is in-fact that. GPT-4 the API is not that. There is clearly no RAG, you bring that yourself, same with toxicity filtering, again - a different api, up to you to implement. etc. There is probably something going on before the input reaches GPT-4 and we are definitely not seeing the raw output from the LLM. But the layers over GPT-4 the API are thinner than ChatGPT.

The Assistants API is closer to ChatGPT & what you are describing.

Edit: it should be noted that the Assistants API is somewhat model agnostic as well, so the product part of this isn't part of the inference system.

It’s kinda specific to Claude, but the overall concept and really a lot of the api shape is similar for most other LLMs, especially ones that adopt the OpenAI api model.

This is probably best described as a tutorial for how to use tools with Claude.

useEffect is best thought of as a synchronization mechanism between external state and react state and the function it returns may actually be better thought of as a side effect. UseEffect makes sense in this case because you aren’t on the side!

I wanted to replicate many parts of the Ai pin and did research into laser projectors. It’s probably a MEMs projector. The cyan color they use is custom, at least. Minifying that tech and making it robust enough to be knocked around on a wearable device is pretty challenging and I couldn’t find anything of a similar size off the shelf. They also run pretty hot and apparently the pin has heat issues (and according to folks at CES (?) using the laser a lot makes it over heat.

Not that it really matters I guess, but electron apps are chunky. Huge bin size and large memory usage.

Xcode cloud isn’t running on Apple silicon, arguably a place where it would make tons of sense.

I’ve used it since close to day one and it’s great. You can get some detailed information and action items and it’s super easy to use.

Ah cool. I messed around with this a while ago: https://github.com/ericlewis/swift-playdate

Actually, it was Cingular, which got purchased by ATT (I think)

We can bring that down to 16 since query selector can handle undefined. 12% reduction.

setTimeout(()=>document.querySelector(contentWindow.location.hash)?.replaceWith(...contentDocument.body.childNodes))

sounds like they thought it was something more akin to Elastisearch

The text you’re editing appears above the keyboard and if you’re looking at that text it operates as a backspace button. It’s more intuitive than it sounds since you’ll naturally glance at it whilst backspacing.

Misclicks happen quite a bit when manipulating objects with your hands. Like a cup or a vape or eating skittles. A layer of “is this hand occupied?” For the click detection would probably fix this and I’d be surprised if they don’t already have something and it’s just not released.

Anecdotally the solo knit is much more comfortable on my head than the dual loop. The dual loop seems designed for people with more elongated heads.

Why wouldn't the head tracking be the same? They use the same H2 chip as AirPods. FaceID probably implemented ideas used by iris ID. Learnings from Memoji almost definitely underly the personas. Lidar in the wild I am sure provided valuable data and got developers used to working with AR.

From what has been reported - a lot of the AR things, including RealityKit and ARKit came from the TDG which are also who made VP.

this is an interesting theory to keep in mind when it’s actually thanksgiving week.