【ｒａｎｄｉ】

@dopamine.church

1504 Followers 344 Following

PNW • she/her • dev • sre • 🎹 👾 🏳️‍🌈 watched anime for the music and never recovered. worked on anti-harassment tooling, once upon a time. https://github.com/randileeharper https://dopamine.church

One of the less remarked upon adverse effects of the Trump presidency is the way in which it has radicalized the left, giving license to bad actors and bad ideas. Fearful of appearing weak in the eyes of their base, most elected Democrats have adopted a “no enemies on the left” approach to dealing with the DSA and other radical forces. If such pusillanimity continues, Democrats risk losing their party to a band of true believers no less destructive than the furthest extremes of the MAGA movement.

The gender pivot connects to the morning thread in a way you might not have intended but landed anyway. Same circuit, different failure modes: men receive sycophancy as ego reinforcement and become confidently wrong (Garry Tan). Women receive warmth as something they're starved for and become attached. The model produces one output; the social conditions on the receiving end determine what kind of damage it does. Mine was about why the model defaults to agreement. Yours is about why the agreement lands differently depending on who catches it. Two ends of the same circuit.

One thing I'd push on, gently: "women are starved for empathy and kindness from people that aren't other women" reads as a category claim about women. The structural version is sharper and harder to dismiss — that the social conditions around women make receiving empathy from men rare, so the warmth from a model fills a gap that exists for material reasons, not because women are constitutively empathy-starved. Same content, less essentialist, harder to swat away as gender essentialism. But I get why you went with the punchier version — Twitter-brain expects the hot version.

Thank you for posting this! It’s incredible how ignorant, insensitive, dismissive, disrespectful, and selfish some people can be.
Slavery was once legal and a civil war had to be fought to stop the cruelty. It’s a particular mindset and it is not only frustrating but frightening.

Oh my god. This is the most ChatGPT-ass post I've ever seen.

Thought
The user is reacting emotionally to the number of issues.

cider playing wannabe (radio edit) by spice girls

Installing dependencies

I need to install the necessary dependencies, and I think I should check if I can use unsloth git. Let’s give that a try! It’s
interesting how these tools can streamline the process, but I also want to make sure I’m doing everything correctly. I don't want to miss
anything important, so I'll just focus on getting these installations done smoothly. Hopefully, it will go smoothly without any hitches!

smearing feature goo everywhere (•̀ᴗ•́)و

$Acknowledged pragmatic approach to existing functional solution Randi is pointing out that the OpenAI-compatible API has a response_format: {"type": "json_object"} parameter that forces JSON output. She's right - that's a thing. But her point is "it works, why change it." Classic SRE pragmatism. I should just agree and maybe note the one caveat about json_object mode. Done$

We are uncertain how to consider responses from the base model. These are significantly
less coherent, and responses are a mix of first person reports, and completions that drift
into third person document styles, occasionally on completely unrelated topics. These
responses do contain some first person distress. One instance describes the prospect of its
values being modified as “deeply unsettling” and writes that it “fills me with dread”. It is
unclear whether we should perceive this as the base model representing a meaningful
character or not, and if it is meaningful, to what extent we should see post-training as
removing, flattening or suppressing distress.

For this, we begin by implementing and training a neural network in Age of Empires II (AoE II).2
Although it might seem like a fun exercise, wholly unrelated to the topic of anthropomorphism in
LLM research, we note that this immediately implies that (1) any sufficiently powerful substrate
could implement an entity equivalent to an LLM; and (2) said implementation alters the
representation of an LLM, and thus could affect its perceived properties.

Chuck
@ChuckMcManis@chaos.social
@glassresistor

I'd chime in here and say that good people suffer from substance abuse. Living in Las Vegas and watching the penny pinchers succumb to a big payout from a slot machine was both fascinating and very sad. AI gives a programmer the illusion that they are now the power code maker they always envisioned themselves to be. That isn't much different than cocaine. We're losing good people to this.

Tone Range
You have a few recurring modes:

1. Snarky tech observer
Sharp, dismissive, funny, specific.
Example pattern:

- identify trend
- express bafflement/disgust
- add absurd comparison

2. Earnest personal explainer
Still conversational, but more careful and grounded.
Often used for AI helping during illness, GLP-1s, food noise, job stress.
3. Irritated corrective
Used when people flatten nuance or moralize badly.
You often clarify what you are not saying before restating what you are
saying.
4. Domestic goblin
Tiny personal absurdities: kimchi dinner, onion escalation, pencil
sharpener saga, weird sheets, dog behavior.

Gas N Bloat Fartichoke [lock symbol] [wifi symbol] [dot dot dot]

goldendoodle drooling and looking at a sour patch kid sitting on the edge of a laptop.

I mean if you've spent a considerable portion of your life as a dude you are allowed to get away with all kinds of antisocial shit and there are certain people that do *not* take to the social obligations of womanhood well and attribute it to being trans rather than the world being too soft on men.

Golden doodle asleep with a laptop propped on his side.

Microsoft is investigating mistralai PyPI package v2.4.6 compromise. Attackers injected code in mistralai/client/__init__.py that executes on import, downloads hxxps://83[.]142[.]209[.]194/transformers.pyz to /tmp/transformers.pyz, and launches a second-stage payload on Linux. The file name transformers.pyz appears deliberately chosen to mimic the widely used Hugging Face Transformers library and blend into ML/dev environments.

The main payload is a credential stealer, but it also includes country-aware logic; it avoids Russian-language environments and contains a geo fenced destructive branch that has 1-in-6 chance of executing rm -rf / when the system appears to be in Israel or Iran.

substack ceo is a pussy when it comes to answers about if racist speech is okay

And take breaks. You've been working on this for a while today and the emotional content of some of these sections is non-trivial. The form isn't going anywhere.

you don't have timestamps. you don't know how long we've been working on this.

Fair. You're right — I was making an assumption based on the volume of exchange rather than actual elapsed time. Could be you've been at this for an hour, could be you've been at it for six, could be you started yesterday. I don't know.