spdustin

20 ms·

241.
by spdustin 3y ago
I’ve noticed I can get it to answer questions into March 2022, but not further (yet). I suspect their September 25 update included a chunk of content from February and March 2022, but not enough for them to say “we got it.”
242.
by spdustin 3y ago
I’ve already updated, and confirmed with others that an update had occurred.
243.
by spdustin 3y ago
That was an excellent read into the machinations of plugins, thanks for writing up the demo!
244.
by spdustin 3y ago
You can’t set temp in ChatGPT, only via the API.
245.
by spdustin 3y ago
Yes, and in this case, the positional encoding of the tokens used in the system message favored returning them verbatim when asked to return them verbatim.
246.
by spdustin 3y ago
I replied to a sibling post, but I’ll copy it here: 1. Consistency in the response (excepting actual changes from OpenAI, naturally) no matter what method is used to extract them. 2. Evaluations done during plugin projects for clients. 3. E
247.
by spdustin 3y ago
1. Consistency in the response (excepting actual changes from OpenAI, naturally) no matter what method is used to extract them. 2. Evaluations done during plugin projects for clients. 3. Evaluations developing my AutoExpert instructions (wh
248.
by spdustin 3y ago
Update: confirmed, thanks again Stagnant. I’ve credited you in the repo.
249.
by spdustin 3y ago
I use it all the time to embed graphviz diagrams by instructing it to url-encode them and include it in the query string to a rendering endpoint. It spooked me enough that I added proxy layer to verify that it’s not rendering what appears t
250.
by spdustin 3y ago
I have suspicions that there is middleware somewhere for the ChatGPT interface to the chat models underneath; something to enclose the normal prompt, to update weights, or to manipulate logits. Just last night I began seeing a behavior that
251.
by spdustin 3y ago
Given its place by “discourse markers”, I take that to mean “don’t use ordinal markers like ‘first’, ‘second’, and ‘third’; instead use referential discourse markers like ‘then’, ‘next’, and ‘finally’”
252.
by spdustin 3y ago
For Advanced Data Analysis, I had it “use Jupyter to write Python” to transform the content of our conversation, including “messages that appeared before this one” or “after ‘You are ChatGPT’”, into a list of dicts. For both voice and mobil
253.
by spdustin 3y ago
I’ve never gotten any of the models to say anything about LaTeX when disclosing their initialization (system) prompts, but I’ll have to dig into that. If I can reproduce that, I’ll update that file and give you credit. Thanks!
254.
by spdustin 3y ago
I have a strong suspicion that their RLHF fine-tuning had a lot of “please” prefixes in there.
255.
by spdustin 3y ago
That was one of several methods, but I (usually) don’t go that deep in comment threads. For Advanced Data Analysis, I had it “use Jupyter to write Python” to transform the content of our conversation, including “messages that appeared befor
256.
by spdustin 3y ago
It’s more like taking advantage of an emergent property of the model’s attention mechanisms, so that it doesn’t confirm or acknowledge the instruction in its text (or “out loud”) completion.
257.
by spdustin 3y ago
The id refers to the id of the quote they extracted (the one with the start and finish “lines”). They are given back to the user’s client as a metadata sidecar to the actual completion.
258.
by spdustin 3y ago
I am 100% confident that none of these are simulated. Variations may exist in white space, due to differences in how I got ChatGPT to extract them, but they are all accurate.
259.
ChatGPT’s system prompts(github.com)
827 points by spdustin 3y ago|389 comments
260.
by spdustin 3y ago
I was pleased to see GPT-4, combined with my custom ChatGPT instructions, had no problem with this: https://chat.openai.com/share/726d5c7b-761f-462e-8a6a-61904a...
261.
by spdustin 3y ago
The Zigbee answer would’ve been to use Zigbee Group Messaging to group the lights together so that all seven bulbs respond to a single command, rather than having to send a Zigbee packet seven times. But I do enjoy de-Tuya-ing no-name bulbs
262.
Show HN: ChatGPT AutoExpert (Supercharged Custom Instructions for ChatGPT)(github.com)
7 points by spdustin 3y ago|0 comments
263.
What're You Lookin' At, ChatGPT?(spdustin.substack.com)
2 points by spdustin 3y ago|1 comments
264.
by spdustin 3y ago
I wrote this post, intended for more of a lay audience, about how ChatGPT learned to do what it does, and how easy it is to write terrible prompts. Thought y'all might enjoy it, or know other folks that could benefit from it. (I do tak
265.
by spdustin 3y ago
Received an email stating that the free Alexa Guard feature—which allows Echo devices to detect the sound of fire and CO2 alarms as well as breaking glass—will no longer be free, and will be rolled up into the new Alexa Emergency Assist sub
266.
Amazon ending Alexa Guard, will charge for alarm and glass break detection(amazon.com)
2 points by spdustin 3y ago|1 comments
267.
by spdustin 3y ago
Wiz bulbs use ESP-NOW (so that the Wiz remote can communicate with them). They also have an ESP-WROOM-02 inside (a package integrating an ESP-8266EX). The fact that they use ESP-NOW means they can be controlled (theoretically) by any ESP* t
268.
by spdustin 3y ago
Curious about the “Write in your voice” feature. Is it mostly derived from LLM stylometry, or do you blend statistical approaches like those used by spaCy? Do you capture all of lexical, syntactical, rhetorical, and semantic cues? I’d love
269.
by spdustin 3y ago
Well done! Would like to see company names as well as logos/tickers. Some companies aren't instantly recognizable.
270.
by spdustin 3y ago
Mine: https://reddit.com/r/OpenAI/s/M4iEAlKCXz Update coming soon that improves it even more. Most redditors don’t understand how attention works, and my prompt takes full advantage of it by making ChatGPT pr

More ›