ukuina

26 ms·

481.
by ukuina 3y ago
Pointless article. > Per the companywide "objective key results" for the new year shared on Thursday, delivering the "world's most advanced, safe, and responsible AI" is Google's number one priority, accordi
482.
by ukuina 3y ago
Don't forget your towel!
483.
by ukuina 3y ago
Inference cost and latency.
484.
by ukuina 3y ago
What would you replace regex with?
485.
Conflict-Free Replicated Spread Sheets(bartoszsypytkowski.com)
1 points by ukuina 3y ago|0 comments
486.
by ukuina 3y ago
Interesting idea but comically unsafe.
487.
by ukuina 3y ago
You could run it through thrice with a different prompt/temperature/model and pick the majority result (or exit with success on the first two passing runs).
488.
by ukuina 3y ago
This is solved with temperature control.
489.
by ukuina 3y ago
The code interpreter probably has access to sqlite for this purpose, as well.
490.
by ukuina 3y ago
This is why I built HackYourNews.com!
491.
by ukuina 3y ago
I apologize. I was trying to respond to inflammatory language ("poison") with similarly hyperbolic terms, and I should know better than to do that. Let me rephrase: Would AI-powered upscaling/downscaling (not a simple determi
492.
by ukuina 3y ago
Won't a simple downsample->upsample be the antidote?
493.
by ukuina 3y ago
Graffiti!
494.
by ukuina 3y ago
I did! The notes are stored in a local DB (SQLite, IIRC) that were not encrypted on disk while the app was running.
495.
by ukuina 3y ago
OneNote sync conflicts effectively destroyed a document for me, with changes splintered across both copies. All my Inking across the copies had unequal offsets making merging impossible. I had to transcribe everything to text manually and t
496.
by ukuina 3y ago
Here's the philosophical conundrum: Do we prioritize learning new things (improve craft) or create value (build stuff)? The first used to lead to the second, but not as much anymore.
497.
by ukuina 3y ago
That might be the joke.
498.
by ukuina 3y ago
> 1.1221 Sometimes the primary tool is not available. An “always-on” secondary should take its place. > 1.1222 Sometimes log from different tools or locations need to be “merged”, “persisted”. These two are the main reasons I've
499.
by ukuina 3y ago
Not as ugly as I thought it would be! With some refinement and contouring to the body of the vehicle, this could make for a fully-untethered city taxi.
500.
by ukuina 3y ago
Google Doc, though performance suffers for image-heavy or long docs. Wish they would include support for Ink.
501.
by ukuina 3y ago
Custom GPTs that do not use uploaded documents for RAG have no moat because the system prompt can always be exfiltrated. Not sure if the uploaded documents can also be leaked, one chunk at a time. There is currently no way to monetize this
502.
by ukuina 3y ago
The delta is the price for disallowing training on chats while retaining all functionality of the web interface and app (e.g., Voice chat).
503.
by ukuina 3y ago
Put this way, $199 seems overpriced!
504.
by ukuina 3y ago
Thank you for the detailed response! Pivoting into arXiv is a good idea. It helps you have focused prompts and templates. A natural progression is aggregation, categorization, and related paper suggestions. Since arXiv has HTML versions of
505.
by ukuina 3y ago
That's just it: It is highly dependent on your inputs, expectations and usecase. My observations below. Mixtral does about average at summarization. Not enough details are presented in the summary. Passable for non-decisionbound workfl
506.
by ukuina 3y ago
Fantastic. Thank you. Any way to "pin" the current time so it acts as a live world clock?
507.
by ukuina 3y ago
I have spent more time than I would like to admit trying to get 3B-34B models working for "serious" usecases like RAG, code comprehension/generation and text summarization, across dozens of "leading" free LLMs, and
508.
by ukuina 3y ago
I visit your site every day. Thank you for creating it and evolving it past simple summaries to show paper details! I recall you were looking to sell it at some point. Was wondering what that process looked like, and why you ended up holdin
509.
by ukuina 3y ago
Gemini Pro seems at par with GPT-3.5-Turbo, and lags behind GPT-4.
510.
Gemini in Reasoning: Unveiling Commonsense in Multimodal Large Language Models(arxiv.org)
66 points by ukuina 3y ago|38 comments

More ›