Those companies buy ads more than they earn revenue from them, Disney being a huge ad buyer. Many ad tech companies existed off the backs of Disney slop. I’d be surprised if those companies care about the ad revenue at all, but rather it’s a way they can reduce their own internal ad spend and create stronger price discrimination so they can profit maximize their higher tiers (how much would they spend to NOT see an ad).

They’d depart from Tijuana I guess.

I was stuck in Eugene over the pandemic, but it was a fine, nice bikeable city. I didn’t want to stay there though, I was a bit too old to make the most of it. In my opinion, Bend is like Denver, but not my cup of tea.

Honestly, I love Portland. I stayed there for many years, as well as Eugene. It’s really rainy and you got to watch out if you have allergies. But otherwise, it’s a great place. I found dating there to be good.

Nice

Blood is our 9th largest export, surpassing trucks

Bad time to be a soy bean farmer, great time to be a soy bean smuggler.

It’s just a semiotics perspective. Try reading Language in Thought and Action (1949) by S.I. Hayakawa, it has basically the same vibes. Chomsky was a linguist first after all.

deleted by creator

This is why I like bimboism/himboism. Society wants us to be smart and you can reject that. It’s okay to be dumb and present yourself as dumb. Honestly, it’s freeing and it’s something I need to exercise more.

I made software I absolute hate in principle but it works so well in practice. It’s all kubernetes (bad), yaml (bad), graphql (bad), and made for vibe coding (bad). I hate myself for spawning this monstrosity into existence but its the shortest path to build the thing I wanted to build for a few years.

I had a geocities when I was about 10 years old where I would put cheats for video games and just random stuff. I would come up with my own game genie code on the NES by changing a few values of existing codes. I found some cool stuff and it was a ton of fun. I had this big book about HTML that I memorized and made my website from scratch. At this point I’m not sure I’d be able to recognize it even if it was on there.

The previous CEO of CrossFit unironically said that to me at a corporate meeting.

No but it does come with fries.

My local Alamo Drafthouse charges $20 for a hotdog.

DSA is an arm of the Chinese Communist Party actually. If you join, they’ll force you to go to China to be brainwashed. You’ll come back talking about the joy of high speed trains and how we can have a bright future if we all worked together. And it’s all completely legal. Careful out there y’all.

Write a list of every reason you why liked her and go over it one by one and remind yourself that you can find that in another person. Works like a charm.

Hey, good job, looks cool. A few observations: It makes sense to aggressively use token caching. It makes sense to improve context accuracy by treating docs as external to the input prompt. Smaller context = more calls, faster CoT, higher llm concurrency. We see a lot of this now, where coding agents aggressively cache documents, and use of tools to do more complicated tasks, e.g. code gen as tool use. I also see more moving away from RAG and more towards using good ole gnu tools like grep as exploration hints. RAG approaches are definitely is more prone to contextual bandit problems. But IMO, balancing exploration with exploitation is fundamentally a reinforcement learning problem rather than a linear flow from recursive llm calls, but it acts a reasonable surrogate model, and our RL algorithms consistently fall short. For example, you could help govern exploration behavior using deep-q or ppo to guide search, e.g. have your llm produce a top-5 exploration s-expression horizon, feed the exploration into the rl algorithm with cosine similarity on vectors + metadata (file size, LOC, etc) as features, then rank the exploration space. This would further reduce token count and increase parallelization in exploration (since top-n results are computed as batch in the llm with only marginal overhead).

Personally, I just wish I had a good search tool that didn’t expose all of the LLM tomfoolery. For example, today I needed to find where a particular sql table was defined in the code since it used some crazy ORM and an initial grep didn’t yield results so I jammed into openai codex and it found it. This was useful to me where github search was useless. I need more context-aware search, and it would be better if I could combine it with all of the other docs and slack convos floating around. Actual coding tasks require a greater level of human interaction where opaque results are less useful.

But again, good job, looks cool.

I bought some cereal from the discount grocery store because it reminded me of cat food. First two ingredients are soy and pea protein, with some flavorings (strawberry). I’m eating it straight from the bag, it’s okay. Thanks for asking.

I used to have a hair tie but then I cut my hair short and eventually lost it. Now I keep everything loose in my pocket like a troglodyte. Occasionally when take off my jacket all my cards will spill all over the floor.