catlifeonmars

Cam shafts predate them.

Let me put it this way. How does the model know what it finds confusing?

I’m sure it works great fwiw, also the use of a hot cache is clever.

That’s not the use case of the tailscale daemon though, which was using getent passwd.

To clarify: I’m not against CGO, just introducing a glibc dependency by default. I would only introduce it when I needed it.

Theory. I guess my question is: why do you think it works so well. What’s the mechanism behind it?

It feels like it would be relatively random across the range of plausible results, but it sounds like you are finding otherwise.

You won’t get grounded results by asking for feedback. The LLM doesn’t know its own internal state so while you may get a plausible and contextually significant answer, it isn’t going to be an accurate one.

Others have pointed out that os/user.Lookup is a platform-independent way to resolve this, but additionally you don’t _need_ to link against glibc to use it.

If you are writing go, you usually want to set CGO_ENABLED=0 by default, to avoid inadvertently introducing nonportable code. In this way, only the pure Go implementations are used and there is no need to link (statically or dynamically) against a libc implementation to compile and run your programs.

“--“ doesn’t work on all versions of getent.

A better fix is to call “getent passwd” with no user controlled arguments and then parse the resulting list. This gets rid of the input sanitization problem entirely.

Do you use git reflog?

What does it mean to split a branch in two?

Looks like a strategy for storing a secret across multiple servers such that no single node compromise will result in the secret being compromised.

My read is that this adds network topology as an input to what is normally just M out of N secret sharing without any consideration for topology.

I just skimmed the paper so take it with a grain of salt.

Hmm this is a problem that has long been solved with code completion in a JSON language server and a valid JSON schema.

By this I mean that you can basically tab-complete your way into a valid JSON document matching a particular schema.

Maybe there is a way to restrict the set of output tokens based on the current position in a JSON schema of a JSON document the model is predicting?

The justification in TFA smell like premature optimization. A nominal bottleneck is just that-nominal. It’s a good thing to keep in your back pocket but you generally don’t want to expend extra effort optimizing something until you have to.

I suspect that the cost of long compilation times (preLLM even) was actually quite high and the author is discounting/excluding those other factors and focusing on LLM feedback loops primarily as their justification.

As an aside short feedback loops are important, but the article ignores one major reason why: humans learn most effectively when the time between action and feedback is reduced because the contents of our working memory degrades over time (take the explanation with a grain of salt). LLM has no such restriction, so the only thing that matters is that the codegen can keep up with the demands from the actual product.

Maybe there were other factors? Maybe they were leaning towards leaving anyway and this influx of cash enabled them to do so? It’s a stupid idea, for sure, but I think the explanation that “people do weird things when faced with a lot of money” is not really all that explanatory.

What is it about json that models struggle with, in your experience? Is it syntax or structure?

TFA reeks of over-sensationalizing. Here is a summary sans hyperbole:

Intranasal Human NSC-Derived EVs Therapy Can Restrain Inflammatory Microglial Transcriptome, and NLRP3 and cGAS-STING Signalling, in Aged Hippocampus[1].

Abstract:

Neuroinflammaging, a moderate, chronic, and sterile inflammation in the hippocampus, contributes to age-related cognitive decline. Neuroinflammaging comprises the activation of the nucleotide-binding domain, leucine-rich repeat family, and pyrin domain-containing 3 (NLRP3) inflammasomes, and the cyclic GMP-AMP synthase (cGAS)-stimulator of interferon genes (STING) pathway that triggers type 1 interferon (IFN-1) signalling. Studies have shown that extracellular vesicles from human induced pluripotent stem cell-derived neural stem cells (hiPSC-NSC-EVs) contain therapeutic miRNAs that can alleviate neuroinflammation. Therefore, this study examined the effects of late middle-aged (18-month-old) male and female C57BL6/J mice receiving two intranasal doses of hiPSC-NSC-EVs on neuroinflammaging in the hippocampus at 20.5 months of age. Compared with animals receiving vehicle treatment, the hippocampus of animals receiving hiPSC-NSC-EVs exhibited reductions in astrocyte hypertrophy, microglial clusters, and oxidative stress, along with elevated expression of antioxidant proteins and genes that maintain mitochondrial respiratory chain integrity. Moreover, hiPSC-NSC-EVs therapy decreased the levels of various proteins involved in the activation of the NLRP3 inflammasome, p38/mitogen-activated protein kinase, cGAS-STING-IFN-1, and Janus kinase and signal transducer and activator of transcription signalling pathways. Furthermore, in vitro assays using genetically engineered RAW cells and hiPSC-NSC-EVs, with or without targeted depletion of specific miRNAs, demonstrated that miRNA-30e-3p and miRNA-181a-5p, both present in hiPSC-NSC-EVs, can significantly inhibit the activation of the NLRP3 inflammasome and the STING pathway, respectively. Additionally, single-cell RNA sequencing conducted 7 days post-treatment revealed that hiPSC-NSC-EVs induce widespread transcriptomic changes in microglia, including increased expression of numerous genes that enhance oxidative phosphorylation and reduced expression of abundant genes that drive multiple proinflammatory signalling pathways. These changes mediated by hiPSC-NSC-EVs were also associated with improved cognitive and memory function. Thus, intranasal hiPSC-NSC-EVs therapy in late middle age can effectively diminish proinflammatory microglial transcriptome and signalling cascades that drive neuroinflammaging in the hippocampus, contributing to better brain function in old age.

[1] https://isevjournals.onlinelibrary.wiley.com/doi/10.1002/jev...

you would probably have to pay for it

_probably_ is doing the heavy lifting here

Is this the one with a hash collision?

You violated the first rule of graphing: label your axes.

a^2 + b^2 = c^2 is perhaps a poor example, since there are several visual proofs that don’t rely on language. I don’t think this negates your overall point fwiw.

Do you have a source for this?

Tokenmaxxing was just a way to force employees to start leveraging AI in a meaningful way.

It was always a temporary thing to transition the employees to a new world.

Trying to understand your justification for rejecting Hanlon’s razor.

Those look like mostly nonsense/trivial findings

What does it mean to “train the AI”?

You think you want that but you don't actually. IF you do sincerely want that, then I sincerely hope you get what you want, but both ways for you and only you, while the rest of us stay isolated as spectators.

Nobody wants this. This is the result of the breakdown of trust in the judicial and democratic processes. GP is just acting the zeitgeist.

Terrorist is often a term used to describe freedom fighters, in order to delegitimize them. Basically this is such a common tactic that it is unsurprising that the two are sometimes conflated.

Git commit?

Wouldn’t it be simpler to use a session token? This complex machinery does nothing but look fancy.

The application secret is redundant if the per-user secret is used.

Also I’m inferring from the article that the author is using symmetric keys (HS256) for their JWTs. In what world can you securely distribute symmetric keys but can’t use an opaque session token?

You don’t even need autonomy to be deadly. Fly by wire is proving to be very effective as a case in point.

These can be both true:

- there are people who are sincerely concerned about model safety who work at OpenAI and Anthropic

- there are people who are using this concern to generate fear to sell a product who work at OpenAI and Anthropic

It makes sense to be cognizant of the apparent conflict of interest and not take things that at face value, especially when there is so much money involved.

You mean the people who have a powerful incentive to lie and exaggerate to sell a product?

Eh, I think you’ll hit diminishing marginal returns over what the SOTA - 1 versions already enable in the hands of motivated professionals.

That ship has already sailed.