Y
HN Search
Hacker News Search
new
|
comments
|
top
|
jobs
spdustin
searching PlanetScale…
1.
▲
2.
▲
3.
▲
4.
▲
5.
▲
6.
▲
15 ms
·
181.
▲
by
spdustin
3y ago
Videos from past Microsoft TechEd conferences, and the 2nd annual SharePoint conference. I had two presentations that did NUMBERS back then, but I should've archived the videos.
182.
▲
by
spdustin
3y ago
Where’d you find the VGA cable? I’ve held on to my MV for years in hopes I could find time to resurrect it. I only had a dock connector, like you. I thought about just using composite and an external video DAC, but I know the thing can take
183.
▲
by
spdustin
3y ago
Happens to me a lot, too. I just switch browsers and it stops.
184.
▲
by
spdustin
3y ago
I have almost 35k uses of mine, so their list is ... lacking. https://chat.openai.com/g/g-LQHhJCXhW-autoexpert-chat
185.
▲
by
spdustin
3y ago
This is what you seek: https://github.com/alexCajas/EmbeddedMqttBroker
186.
▲
by
spdustin
3y ago
Pretty much everyone running a Christmas light show on their house is sequencing it with this, and playing it back on a raspberry pi or beaglebone black, sending signals to a bunch of ESP32s to control their lights. And many of them buy the
187.
▲
by
spdustin
3y ago
Okay, that one's handy as hell. Thanks for sharing.
188.
▲
by
spdustin
3y ago
I'm upvoting this because learning these devices can be wildly useful when prompting LLMs for content generation . More so, if you use these devices in stylometric analysis of your own writing, as it makes "style transfer" a
189.
▲
by
spdustin
3y ago
I really want to be excited by the 2.1 updates, but as a Bedrock user, I'm still unable to use "system" prompts. AWS' Bedrock's `invoke_model` still errors out if the prompt doesn't start with "\n\nHuman:
190.
▲
Google's NotebookLM (RAG-based chat) now available in the US
(blog.google)
2 points
by
spdustin
3y ago
|
0 comments
191.
▲
by
spdustin
3y ago
> "We took steps to protect our users by blocking techniques that exploit fake credentials in order to gain access to iMessage." They never addressed that part of Apple's statement. I don't see how they can survive wi
192.
▲
by
spdustin
3y ago
Why not make the CSS dynamic? Or even manipulate the SVG itself in script? D3 has been doing the latter for a long time.
193.
▲
by
spdustin
3y ago
Combination of repetition penalty and overwhelming the attention mechanism, depending on how you're using it (ChatGPT and ChatGPT/API, respectively)
194.
▲
by
spdustin
3y ago
Please add an option to delete account. PayPal is the only free one? Moving on.
195.
▲
by
spdustin
3y ago
I’ve got them all here, including the original Custom GPTs and the Custom GPT Builder: https://github.com/spdustin/ChatGPT-AutoExpert/blob/main/Sys...
196.
▲
by
spdustin
3y ago
I requested access through Bedrock and had it minutes later. It's an automated process.
197.
▲
by
spdustin
3y ago
A DoS is a theft of computer resources. Since that's specifically a crime in most jurisdictions, it's not protected speech.
198.
▲
by
spdustin
3y ago
Unless you're a member of a protected class, the numbered act violates a law, or the numbered act violates contractual terms: 13.
199.
▲
by
spdustin
3y ago
CodeWhisperer surprisingly effective as a CoPilot in VSCode—quick, too—and this looks quite useful. Especially the text-to-shell tool, looks frictionless.
200.
▲
by
spdustin
3y ago
Do you have plans to support Bedeock? In a hosted environment, I recognize it's a little more...involved. But still, I'm curious.
201.
▲
by
spdustin
3y ago
Simon, I’ve got the full Custom GPT Builder prompt here: https://github.com/spdustin/ChatGPT-AutoExpert/blob/main/_sy...
202.
▲
by
spdustin
3y ago
Technically, it is (it’s gpt-4-gizmo). There’s a chance that had some fine-tuning for understanding the presence of the just-in-time function calls for actions. (Names for Actions in Custom GPTs are tagged with “jit”)
203.
▲
by
spdustin
3y ago
It’s worth mentioning that it’s not entirely classic. It’s still using the 32k context turbo model.
204.
▲
by
spdustin
3y ago
POST operations don’t get “always allow” because they might mutate a resource.
205.
▲
by
spdustin
3y ago
10 for GPTs, 20 for Assistants
206.
▲
by
spdustin
3y ago
The sharding done for turbo models has almost certainly affected the attention mechanisms. It makes me think the sharding was done at the layer level.
207.
▲
by
spdustin
3y ago
The citations are built to reference the ID of the quote object in the metadata returned by the `quote_lines` function. I have been able to get them to point elsewhere, but not in the GPT itself; only with a userscript that intercepts the f
208.
▲
by
spdustin
3y ago
Yes, it's another Custom GPT, but this one's different. It maximizes the available instruction space, and integrates with APIs (currently arXiv, I'll add more tomorrow) to help you find papers you're looking for. Here&#x
209.
▲
AutoExpert (Academic) GPT for analyzing scholarly papers
(chat.openai.com)
1 points
by
spdustin
3y ago
|
2 comments
210.
▲
by
spdustin
3y ago
I share the ones I use every day at https://github.com/spdustin/ChatGPT-AutoExpert
More ›