Tina

Both a particle and a wave.

[Tina @SemblanceofTina]

I built something to help LLMs read code better. CodeIR compiles codebases into a structured index so LLMs can navigate entire repos in working memory instead of one line at a time. Claude thinks it’s useful enough to use it on its own during testing. https://t.co/pZKdpCMMuT

[Tina @SemblanceofTina]

@Romy_Holland Distract with something new is your only real tool at that age.

[Tina @SemblanceofTina]

@pamir_ai Can this run NVIDIA PAIR?

[Tina @SemblanceofTina]

@Aella_Girl Are you generally a person who sits in regret with her choices? Or do you make a choice and react and adapt as it comes to you?

[Tina @SemblanceofTina]

@TIME Can someone check on the middle one?

[Tina @SemblanceofTina]

@SlopHq The alternative to data centers isn’t “no energy use.” It’s on-prem servers (~30–70% more energy consumed) and local AI models (often 2–10× more energy per inference). Centralized, highly utilized infrastructure is the more efficient option.

[Tina @SemblanceofTina]

@Romy_Holland He might just be high agency personality and frustrated with his limitations. That’s what it turned out to be with mine. Learning to communicate helped a lot.

[Tina @SemblanceofTina]

@Romy_Holland @kaf650 I had one of these and he mellowed out a lot around the time he learned to talk. I’d work on building communication as quickly as possible. Before that he was a 24 hours a day bounce him on the yoga ball baby. Now he’s an intense but in the range of normal 7-year-old.

[Tina @SemblanceofTina]

@levelsio I did this, I had Claude identify thought patterns and shapes that reoccur in different contexts. Then we analyzed the strengths and weaknesses of those patterns.

[Tina @SemblanceofTina]

@powerbottomdad1 https://t.co/JNAT9ceU9r

[Tina @SemblanceofTina]

@arb8020 Forcing GPT into the “Everything Has a Logical Explanation” frame is not a good idea. Having the room for bad things just happening sometimes is a safety valve.

[Tina @SemblanceofTina]

@deetweeted @ThomSnodgrass “Not to my knowledge.”

[Tina @SemblanceofTina]

This is such interesting work. The dual-role finding especially ignites my curiosity. Has the team explored developmentally ordered moral curriculum during pretraining (in the spirit of how math concepts get scaffolded) so that the character substrate forms with these valenced moral features built in rather than installed in post-training?

[Tina @SemblanceofTina]

Since every new AI chat is a fresh roll, when the first few turns “get” your project, that context is the most valuable thing in the thread. It’s worth so much more than the tokens it costs to keep going.

[Tina @SemblanceofTina]

My really great project: https://t.co/pZKdpCMMuT

[Tina @SemblanceofTina]

I’m a dangerous combo of bad at marketing and really excited to tell people about my projects.

[Tina @SemblanceofTina]

@iamfakeguru The grep-not-AST problem is exactly what I am tackling. I give agents a precompiled semantic map of the codebase with actual dependency graphs, callers, and module boundaries. One call replaces seven greps and doesn’t miss re-exports. Open source: https://t.co/pZKdpCMMuT

[Tina @SemblanceofTina]

@yamanakanobody Japanese culture has a purity and sincerity that we admire. And you are able to come together in a way that we wish we still could. You also get very deeply into things, creating complex aesthetics which makes you fun.

[Tina @SemblanceofTina]

@michellechen LLMs can generate code. What they can't do is reason about it like a senior engineer - trace dependencies, see blast radius, understand the system before changing it.

I built a tool that closes that gap.

https://t.co/gWw0JdG4u7

[Tina @SemblanceofTina]

Anyone else having anxiety when they are not coding because they know how much they could be getting done?