tippytippytango

83 ms·

31.
by tippytippytango 2y ago
I’d bet the hypothesis was already in the training data. Someone probably suggested it in the future work section of some other paper.
32.
by tippytippytango 2y ago
Don’t ignore any question is the strategy I’ve found. The problem is that some questions fall just below the conscious threshold. Meditation seems to help dredge up the questions and take them seriously.
33.
by tippytippytango 2y ago
The people are intuiting this. I think the next election cycle will see a left wing strongman put in. That one will do damage after cleaning up the damage from the current one. So we’ll yoyo back and forth between strongmen to get shit done
34.
by tippytippytango 2y ago
I love that the kids are rediscovering things like landlines and baggy pants.
35.
by tippytippytango 2y ago
Exactly, they want to automate the most rewarding part that we don’t need help with… plus I don’t believe they’ve solved the problem of LLMs generating trite ideas.
36.
by tippytippytango 2y ago
If I give you a hard drive full of photos, you could compress them, use the excess hard drive space for your algorithm, then uncompress the photos and give a full hard drive back to me. Is that’s what’s going on here? Basically it’s exploit
37.
by tippytippytango 2y ago
I like to play white noise or coffee shop ambiance on my noise canceling headphones, I wonder if that helps?
38.
by tippytippytango 2y ago
Senior devs have a decade long reinforcement learning loop with the marketplace. That will be a massive advantage until they start RL on agents against the same environment.
39.
by tippytippytango 2y ago
I have a feeling that "Apple is late, but does it right" is an excuse. They are dealing with a tradeoff from being a deeply integrated and polished system. Being in control of everything is nice for creating a unified experience.
40.
by tippytippytango 2y ago
Good to know openai knows the frustration of trying to argue with their RL based models as well.
41.
by tippytippytango 2y ago
What I learned is you can’t outsource expertise to an LLM, after many similar experiences to OP. Don’t ask it for advice, ask it to teach you so you can make decisions like this on your own. Preferably ground questions with excerpts from hu
42.
by tippytippytango 2y ago
You can’t make up for a fundamental lack of intelligence in the base models with a bunch of hand coded heuristics.
43.
by tippytippytango 2y ago
This is not a concern when you are responsible for real results. If you aren’t responsible for real results you can pass off the good rhetoric of these models as an “answer”. But when you need results you realize most answers they give are
44.
by tippytippytango 2y ago
This is self correcting. Code of this quality won't let you ship things. You are forced to understand the last 20%-30% of details the LLM can't help you with to pass all your tests. But, it also turns out, to understand the 20% of
45.
by tippytippytango 2y ago
The bitter lesson is a powerful heuristic, but its adherents treat it as dogma and wield it as a club to win arguments.
46.
by tippytippytango 2y ago
If you are rich you have the ability to become a scientist without worrying about money. That is an absurd privilege that I am personally working towards. To contribute to science is one of the most enduring things you can do for humanity.
47.
by tippytippytango 2y ago
Evolution just pushes straight to master.
48.
by tippytippytango 2y ago
Yes! Markets work better when deceptive price signals are not inhibiting price discovery. When deception is not rewarded.
49.
by tippytippytango 2y ago
An LLM is only a threat if writing code is the hardest part of your job.
50.
by tippytippytango 2y ago
Fintech needs a lot more regulation if people are having to worrying about this kind of nuance to engage with the business.
51.
by tippytippytango 2y ago
What is in Intel’s culture, vision and values that is needed in the current marketplace? What is unique and valuable about their perspective that no one else can provide?
52.
by tippytippytango 2y ago
You break through the 70% barrier by writing detailed spec and architecture documents. Also tests that define behavior. Those go in every request as you build. Don’t ask an LLM to read your mind.
53.
by tippytippytango 2y ago
It’s like hiring an assistant. You could hire one for 60k/year. But you wouldn’t do it unless you knew how the assistant could help you make more than 60k per year. If you don’t know what to do with an employee then don’t hire them. If
54.
by tippytippytango 2y ago
“But it makes mistakes sometimes!” Cool bro, then don’t use it. Don’t bother spending any time thinking about how to create error correction processes, like any business does to check their employees. Yes, something that isn’t perfect is wo
55.
by tippytippytango 2y ago
There’s only so much you can do when you train on the data instead of the processes that created that data.
56.
by tippytippytango 2y ago
A cautionary tale for Nvidia to not become the next Intel.
57.
by tippytippytango 2y ago
These days I usually only search google for an exact phrase or title I got from chatgpt to check if it hallucinated.
58.
by tippytippytango 2y ago
Given kids use these tools so much I wonder if there’s a reverse clever hans effect where kids imitate the chatgpt style for legitimate essays. Either way, this is a giant lawsuit waiting to happen. Schools need to ban these tools asap. The
59.
by tippytippytango 2y ago
This is what we really need AI regulation for. The accuracy rates should be advertised in a standard format like a nutritional label. People purchasing the systems on public dollars should be required to define a good plan for false positiv
60.
by tippytippytango 2y ago
I tried these with much excitement. Unfortunately they give me immediate eye strain and headaches. The optics need a tremendous amount of work. 1080p is also not enough. But some people love them. It’s a product you just have to try and see

More ›