Jim Su

@CliffordAsness Even dumber tweet

@0xJirachi is this dumpling time?

best xiaolongbao in sf hands down (gatekeeping)

to the moon

your team needs custom claude code verbs

fire a16z happy hour @JacobZietek

New bar for product market fit: when somebody wants to use your product so bad they’re willing to invoke the Defense Production Act of 1950 to keep using it

@patrikkml me prompting my level 20 bytox

vibecoding sum bullshit rn

@evisdrenova why does it not exist in shenzhen?

People will say proximity to investors etc and that's all true. But as someone who has lived in Boston, Chicago, SF and spent considerable time in NYC and Austin, the biggest benefit is actually intangible. It's the energy. The vibes you get from the billboards, the number of people you see coding in the cafes, the brief technical conversations you overhear in the street, the logos on people's backpacks, the number of people you see taking walks during the day deep in thought. You know that down the street someone is working hard on a new idea, trying to create something from nothing, even if you don't know who it is. That energy, at that level of concentration in one industry, does not exist anywhere else in the world. And it drives you to think bigger and move faster.

during code review, the intern said my function had a bug. i told him: bro - you clearly have no idea what you are talking about. i have shipped more features than anyone on this team (combined). then i approved my own pr and mass deployed to prod.

@raaleh jimsu.creator-spring.com/listing/openai…

Put this on a shirt

@burkov GPT-4 is also 1T+ parameters and cost $100M+ to train. The scale is hard to compete with. wired.com/story/openai-c…

GPT-4 wins over open LLMs for one simple reason: while the open-source community throws random finetuning datasets on pretrained models, OpenAI sees what queries real users submit and writes high-quality answers for finetuning.

@jxmnop How about a bigram model? Would essentially be one nn.Embedding layer.

what’s the simplest neural language model i can code for my little class? maybe just an MLP on some word embeddings? doing an RNN feels like it could be overkill

@theojaffee For people, it’s not only possible but also common. It’s not the smartest person who becomes the president, it’s who the country believes can lead.

A nuance that often gets lost in the AI debate is that it *is* possible to actively control something smarter than you, and not just in the sense of aligning it once with your values and letting it loose. The level of actual control diminishes as the system gets smarter, but it’s still there. A trivial example is that 100 IQ guards can keep 160 IQ Ted Kaczynski imprisoned, a nontrivial example is that your very stupid limbic system has a great deal of control over your very smart neocortex

@theojaffee Ezra Pound Vin Diesel Barack Obama

What are some examples of people with first and last names that are each sufficient to identify them? - Elon Musk - Eliezer Yudkowsky Who else?

@maanav Since the value proposition of art/music is largely entertainment, I think less demand for them => harder to succeed. These are also the most subjective fields, so it's hard to measure progress toward success. E.g., how do you ensure that your next song is better than your last?

was just thinking about why it’s so much harder for artists/musicians to succeed compared to other careers maybe because they need you to spend not just your money, but also your time/attention on them?