jerpint

24 ms·

331.
by jerpint 3y ago
What model are you using in the backend? It’s pretty fast to respond. Very nice interface
332.
by jerpint 3y ago
For me, the killer vim feature, apart from navigation of course, is macros
333.
by jerpint 3y ago
Should be “and learn encryption” not “or”
334.
by jerpint 3y ago
It can reinforce its own biases for one
335.
by jerpint 3y ago
In hindsight it’s not surprising but when you’re paying turkers you expect humans to be doing the work, good to know it’s no longer reliable either
336.
by jerpint 3y ago
For every hype there’s an equal and opposite reaction to the hype. Proposing ML to sort comments is a totally valid suggestion, but might be difficult to get right
337.
by jerpint 3y ago
For me GPT allows me to explore more ideas quickly, and can help you learn languages more efficiently. More importantly, you don’t have to use it
338.
by jerpint 3y ago
Can’t wait for soulless 24/7 grocery store robot music /s
339.
by jerpint 3y ago
These tricks might not but certainly others will. GPT4 was shown to loose reasoning abilities the more it got RLHF’d
340.
by jerpint 3y ago
What LLM are you using in the backend? Is this an openAI model or a model trained by wolfram?
341.
by jerpint 3y ago
A lot of industry solutions will require more than simple prompt engineering or running the QuickStart from a tutorial. Understanding and grasping the fundamentals of these models will be a huge asset, and CS degrees could definitely help w
342.
by jerpint 3y ago
I personally don’t use copilot. Sometimes I ask chatGPT for help on small, atomic functions. It has saved me lots of time, and the fact that I have to explicitly copy+paste makes me more aware of the code I’m potentially copying over
343.
by jerpint 3y ago
Temperature = 0 will give deterministic results, but might not be as “creative”. Also it’s not enough to guarantee determinism , hardware executing the LLM can lead to different results as well
344.
by jerpint 3y ago
> Each chatbot has unlimited memory capacity, enabling seamless interaction with large files such as a 400-page PDF. How is this possible? Do they do fancy tricks at inference?
345.
by jerpint 3y ago
another approach is HyDE, where you ask the LLM to come up with a plausible (but likely wrong) answer, and use the embedding of the wrong answer to find the appropriate chunk, pretty clever
346.
by jerpint 3y ago
Does this offer any advantage over gradio? One big issue I have with gradio is that « simple » computations happen in the backend and make the UX very slow on deployment
347.
by jerpint 3y ago
One « best practice » completely ignored by this document is how to ensure non-stochastic results (eg temperature=0), and better yet how to be « sure » which version of chatGPT you’re using (currently no way of knowing). I wish they would g
348.
by jerpint 3y ago
Models never « see » anything to begin with, it’s all matrices. And since we ditched convents, locality doesn’t even matter anymore (almost). RGB is just convenient for humans, nothing says it’s optimal for deep learning
349.
by jerpint 3y ago
Even if we believe the lack of agenda of openAI there is no way to ever certify chatGPT doesn’t have its own (stochastic) agenda
350.
by jerpint 3y ago
This is a very important detail. Temperature > 0 means a different result every time
351.
by jerpint 3y ago
We are biased , as HN users, to prefer text-based, minimal, UIs that lead to (mostly) insightful conversations. Reddit on the other hand caters to a hivemind type mentality that loves things like emojis, karma, dramatic posts, etc. The curr
352.
by jerpint 3y ago
If you go to the front page without logging in, you’ll quickly notice the target demographic is teens/young adults. The redesign makes a lot more sense with that context
353.
by jerpint 3y ago
maybe attention IS all you need
354.
by jerpint 3y ago
When I use my laptop on the tray, if the person in front reclines their seat quickly and I’m not paying attention, there’s a good chance they’ll crack my screen
355.
by jerpint 3y ago
tmux but a browser
356.
by jerpint 3y ago
LLaMa was not fine tuned on human interactions , so it shouldn’t be subject to the same effect, but it also means it’s not nearly as good at having conversations. It’s much better at completing sentences
357.
by jerpint 3y ago
Deepmind was behind this kind of thinking years ago with Perceiver models, but meta is crushing it with high quality publications lately
358.
by jerpint 3y ago
This sounds like a GPT summary, useful nonetheless
359.
by jerpint 3y ago
Couldn’t find thumb up or down on your picker
360.
by jerpint 3y ago
Ouch

More ›