drusepth

23 ms·

331.
by drusepth 4y ago
I think that's kind of a bigger issue with Dall-E: they just sat in the middle of the two consumer extremes, without a differentiating feature themselves. Midjourney ate away at them from the quality highground while Stable Diffusion b
332.
by drusepth 4y ago
Given how strict OpenAI has been about what you can do with their API in the past and how hard it was to get some legitimate apps through approval, I would imagine they'd just shut this competitor's API access down.
333.
by drusepth 4y ago
I hadn't seen that! It's a great step in the right direction, although as I understand it it's more of an opt-out system for LAION rather than its own new dataset, correct?
334.
Ask HN: Is anyone working on an opt-in alternative to the LAION image dataset?
2 points by drusepth 4y ago|14 comments
335.
by drusepth 4y ago
As an off-topic side note, I saw this trending on Twitter and just assumed it was the latest conspiracy on the site. Seeing it on HN lends it a lot more credibility. I used to watch Twitter for a source of breaking news. It's a shame t
336.
by drusepth 4y ago
Geek, nerd, and dork can all be used positively/affectionately. I know I've called myself and friends all of the above on many occasions.
337.
by drusepth 4y ago
I haven't answered my phone except for family members for coming on 3-4 years now and still get 3-4 spam voicemails a day. That number goes way up when I start screening the calls though, so maybe there is something to answering vs not
338.
by drusepth 4y ago
I don't particularly want to harm advertisers; I'm just interested in the proposed security benefits from OP. This does seem like a realistic middleground though. Thanks for the suggestion.
339.
by drusepth 4y ago
Does anyone have any adblockers they recommend that still show "safe" ads (e.g. non-malware) by default, without having to whitelist every site? I'd be open to the security benefits of an adblocker if I could still passively
340.
AI, Writing, and Publishing(medium.com)
2 points by drusepth 4y ago|1 comments
341.
by drusepth 4y ago
Definitely won't be going back to a physical office, but I understand some people prefer it and/or struggled with working remotely. A little sad that some companies are moving to a limited-hybrid approach instead of making either
342.
by drusepth 4y ago
YouTube has forever had a moderation problem. What some people call censorship, others call moderating. I'm glad the comments/videos have improved significantly in quality the last ~year, due to what I assume is being attributed t
343.
by drusepth 4y ago
>Still, I wouldn't want an employee to talk back to a rude customer. This is an interesting take, and might be the crux of why this issue seems divisive (wrt how Bing should respond to abuse). Many of the screenshots I've seen
344.
by drusepth 4y ago
There's a big difference between tech being able to last 1.5 years and being guaranteed to last 1.5 years. The latter costs more upfront to smaller players, even if the quality of the tech is otherwise identical.
345.
by drusepth 4y ago
Building out the search a little more to support exact matches would also be super useful in this flow. For example, I've been on several podcasts talking about Notebook.ai, but searching for the name also matches "notebook",
346.
by drusepth 4y ago
I'm not sure human connection is an equal comparison with offices. There's plenty of human connection, both in person and not, to be had with coworkers without centralized offices.
347.
by drusepth 4y ago
It's from prompt reverse engineering and therefore may not be entirely correct, but was consistently reproducable for a time even with ChatGPT's high temperature values. A separate RE process also produced the following prompt, wh
348.
by drusepth 4y ago
The current date is given to the model in the leaked initial prompt that each ChatGPT conversation starts with. "You are ChatGPT, a large language model trained by OpenAI. Respond conversationally. Do not answer as the us
349.
by drusepth 4y ago
Technically, a scientific weather balloon could also be classified as spying on our weather conditions. :)
350.
by drusepth 4y ago
Definitely an interesting topic. I actually went and plugged a bunch of my stories/poetry into the new OpenAI human/ai classifier to see what it spit out and it all came back human-written, so at least there's that. :) I see
351.
by drusepth 4y ago
Midjourney has kind of replaced Pinterest (and, honestly, some forms of reading) for me. I use it daily to generate concept art for whatever writing projects I'm working on, which keeps me inspired and working, even if the generated im
352.
by drusepth 4y ago
Absolutely. I built a super simple editor in rails 2 years ago on GPT3 [1] that simply pulls the most recent N words in your document as context and tries three times to complete the next paragraph for you, and just inserts whichever comple
353.
by drusepth 4y ago
I'd hazard a guess that they're gonna start cracking down hard on unofficial API usage, and restrict the subscription to just their web UI. The fact that they're also offering a ChatGPT API soon seems to reinforce that dualit
354.
by drusepth 4y ago
The link we're all commenting on has a waitlist form for the ChatGPT API, FWIW.
355.
by drusepth 4y ago
Upon closer inspection, they have a lot of weights regarding "yellowness" [0], and several of them say "based on Toloka", which is an antifraud service that assigns colors to violation levels [1]. So my suspicion above i
356.
by drusepth 4y ago
It's been a long time since I've read it, but I vaguely remember a write-up many years ago (before we used ML for NSFW detection) about how either yellow images would often produce false positives in popular detection algos, and&#
357.
by drusepth 4y ago
I have a sneaking suspicion the "moat" of AI models will be in the data used to fine-tune them. Prompts are, as you alluded, inherently impossible to fully secure and playing cat and mouse with all the ways they can be compromised
358.
by drusepth 4y ago
I always love apps like this. Would love to try it out when it's available on other platforms someday.
359.
by drusepth 4y ago
I think there's large segments of many podcasts that'd be nice to just get through, where comprehension isn't inherently necessary. :)
360.
by drusepth 4y ago
I wanted to like podcasts, but never found one that wasn't either incredibly boring or incredibly grating to listen to. Maybe I'm not the target audience, but I'm rarely if ever interested in the speaker of whatever informati

More ›