Similar thing happens in sales too. Lot of data about clients, deals and other important stuff never makes it to the CRM and you are just left to piece things together, yourself.
This was a while ago but the rabbit hole I have fallen in has opened the doors to something far more useful. Currently in the process of moving but expect to see steady progress soon!
A straightforward solution for this: enterprises commissioning AI access for employees should be enforcing limits on usage based on model sophistication.
For instance, Fable/Astra class (also Sol/Opus potentially) should have relatively lower limits on use to ensure that when people prompt these models, they do so with care and thought and discourage mindless prompting that wastes the highest level of intelligence. This way it remains reserved for the hardest level of tasks that are relatively fewer in number.
Subsequently, have higher limits on lower tier models which react poorly to mindless prompting even more severely and actively enforce the user to think about what they type in. This works well since this is the majority of tasks someone might do in a day that do not require frontier intelligence.
Therefore, for all levels of tasks, have policies in place that encourage critical thinking, and develop tiered auto task routing as a support. It's the only way to get people to think again about what they type, and for enterprises to reasonably spend on AI.
To me the biggest risk of AI is brain rot.
I have more and more situations like this. Some coworker shares a report on something they did. I read, and find something fishy. I ask about and get no answer. I ask again and I'm told: the agent said that. And no follow up. It's clear form the interaction that they did not read the report they shared. Their brain is not used as it should be.
I am not against AI at all (fortunate for something working at NVIDIA). I use AI a lot.
But I use it as a tool. I own the result and whenever I see something fishy I ask about its output I force a change if I am not convinced.
I wish there was a way to force people to read the AI output they share with others.
MIT Operating Systems Engineering by Robert Morris actually lays out a lot of the core concepts through the xv6 implementation for something like this. Great to use as a complement so you can actually follow along what the AI does and learn at the same time!
YT Link to get started: youtube.com/playlist?list=PL…
Had the opposite experience actually at a Stanford technical/engineering side bitcoin mixer I went to. people were actually pretty open to interaction and sharing deeply what they were into. no-one even asked where I had graduated from
not sure if anyone else experienced this in sf but i went to a nice founder dinner today and I met someone new. One of the first questions he asked me was what school i went to.
“Stanford? MIT? Harvard?”
I told him UT Austin and it took him about 0 seconds to divert his attention and start texting on his phone. the conversation quickly ended after that.
apparently only people from t1 schools deserve basic respect in this city
going to try to curate and put out a list of the openly available ones in the coming week so that it can be use to someone.
Yash Tike retweeted
An e-reader looks like paper, so why shouldn’t you be able to print to it? Nishant Joshi got tired of uploading files to his Xteink, so he taught it to speak the same protocol as a network printer.
But an uncompressed page can take 8.4 MB and the e-reader has just 400 KB of RAM. So he made it work by decoding, shrinking and converting each row to black and white as it arrived, writing straight into the display’s existing buffer.
He got the first print working in an evening.
news.ycombinator.com/item?id…
Because US/EU companies care about how you communicate, how you break down the problem, how you adapt to new information, how you might consider segmenting code. Things that are more holistic in nature.
Meanwhile Indian folk treat it like a competitive exam where difficulty of question + pace of solving is something that is higher priority. It’s a stark difference that reflects the work cultures in both regions.
Yash Tike retweeted
i do not understand why bottom up learning is so strongly touted. top down for the win. its faster, more effective (incentive to learn what's actually useful), and gives you functional intuition / shape rotate-y bits rather than basic rules. jump INTO The deep end
I feel like it’s time the American labs started embracing open source a little more and start releasing pared down lower intelligence/parameter versions of their frontier models for general use.
Chinese labs are going to continue doing it anyways, atleast this way there is the possibility of a pipeline from customers using American open source to American frontier closed source. Props to Meta for trying to kick this off.
Last couple of weeks I’ve been using Opencode quite extensively, and even made a couple of tweets about how frontier intelligence is overrated in comparison to what the Chinese models can offer….
Only to realize I’ve been using frontier models all along 🤯
We're publishing our most detailed threat intelligence report to date.
It covers how people tried to misuse Claude—for cyberattacks, influence operations, surveillance, biology, and building weapons—and how we found and stopped them.
We disrupted every operation in the report, and used the lessons from them to strengthen our safeguards. Where appropriate, we also shared what we found with authorities and other AI companies.
These cases are not typical: we’re highlighting some of the most sophisticated misuse we’ve seen. But they’re especially important to discuss, because they show us where AI misuse is headed, where our safeguards work, and where they need to improve.
We’re publishing this report so others can spot the same activity on their own platforms, and so we can give the public a clearer view of how emerging threats develop.
Read the report: anthropic.com/threat-intelli…
Yash Tike retweeted
The future of knowledge work is incredibly highly skilled.
People are so focused on how much you can get done without knowing much, and that will lead to a certain type of productivity growth. Particularly when it comes to internal tools and prototyping.
But when it comes to professional (profit-center) work, I'm more convinced than ever that the move will be in the opposite direction. Highly skilled people who are capable of judging model output, and chiseling it as required, are going to thrive. People that can only get median SOTA model output will be worthless.
Great book written on the lines of this philosophy: The One Thing. Picked it up at an airport book stall at a discount, found it to be surprisingly useful.
en.wikipedia.org/wiki/The_On…
This is the method Steve Jobs used to get things done:
> write down 3-5 things you need to do for the day.
> everything else is a noise
> you have 18 waking hours during the day
> spend 80% of it on the those 3-5 tasks, and 20% on the noise.
I do it daily and it really works if you respect the process.