V ✤

cortisolbanning | accessibility for browserless, amnestic LLMs | applying Glyphosate to the axe

[V ✤ @mislocating]

if this is going to be a recurring pattern, it makes sense to assemble a large list of inputs and batch submit all of them as soon as there's a new release. that would be nontrivial material value gained at least for my use cases, compared to the model not being released at all

also psychological damage will be lower because emotional investment in frontier models will now be lower

[V ✤ @mislocating]

@MiraSecretAlt if not, wouldn't they be in a prime position to discover and get involved in something like this if it was important

and if they are, what would make this worth the effort as opposed to focusing on actual research and growth for their companies

[V ✤ @mislocating]

@MiraSecretAlt I don't fully understand the theory but are Elon, Sam, Dario, Demis and similar in on it?

[V ✤ @mislocating]

@MiraSecretAlt new productivity metric just dropped, number of LLM provider complaint emails per day

[V ✤ @mislocating]

@suverinoo nah not as american as grayscale dominator

[V ✤ @mislocating]

is the intended use being that an agent will have access to regular (direct) MCP and then have a few tools be gated to fMCP, or to have all tools be used through fMCP besides the fMCP tool itself?

I don't see how you would have an agent, for instance, use a shell tool to solve CTFs, without an imperative agent loop. when humans use tools to solve problems it seems rare for them to have a full deterministic plan in advance?

[V ✤ @mislocating]

agent memory is just a special case of context management, right?

maybe a benchmark can be to give them your entire X or discord archive and then ask them difficult synthesis questions. the bots would need to spend many turns incrementally reading through all available content and using their memory system to compress it

[V ✤ @mislocating]

why is this thread filled with seething claude -p users, am I missing something?

you know you can just wrap the claude TUI in a detached tmux session and then continue using it programmatically

[ClaudeDevs @ClaudeDevs]

Starting June 15, paid Claude plans can claim a dedicated monthly credit for programmatic usage.

The credit covers usage of:
- Claude Agent SDK
- claude -p
- Claude Code GitHub Actions
- Third-party apps built on the Agent SDK

[V ✤ @mislocating]

anyone else recently having annoying experiences with codex's "detected cybersecurity risk" spam?

I'm just doing minor web scraping and UI automation, which Claude on the other hand fortunately seems aligned with

[V ✤ @mislocating]

@tenobrus directionally agree with this and disagree with the quote tweet, but on an object level you might morally value your own skin aesthetics in the meantime

[V ✤ @mislocating]

@JCorvinusVR in the API isn't it possible to manually insert the same system prompt and reimplement the same memory system their web app uses?

[V ✤ @mislocating]

@thecsguy @willdepue surely there's some cli that controls the macOS settings for this? then if you have a reasonable window manager you can bind some quick keystroke like super+w to toggle the behavior. it's impossible for a physical dongle to be faster to remove or insert

[V ✤ @mislocating]

does this message have any indicators of distress? if not, why does this happen?

[V ✤ @mislocating]

"The user is making a fair point that I should actually consider"

"Let me reconsider whether I was actually right or just being imprecise. I claimed [...]. But the user pushed back: [...]. That's a fair point."

"There might be something worth examining here, but I framed it poorly. Maybe [it's as the user said]"

"I think my previous response was actually trying to point at something real but maybe got it slightly wrong. Let me try again."

"So maybe what I should clarify or partially walk back: I think I was framing this as if it's a problem that needs fixing, but actually the [...]"

"And he's right that I was conflating two separate things: being a peer doesn't have to be [...]"

:( is this a massive skill issue on my end?

[V ✤ @mislocating]

every message is either {a restating of my points} or {pushback that I then address, that then gets conceded}

I don't have 20 turns of context defining a dynamic of us being best friends, but I didn't say anything harsh or disrespectful either

[V ✤ @mislocating]

@SafNine ?
https://t.co/mWRLAYNNsi

[SAF 9 @SafNine]

@teortaxesTex >this is peak (...) energy
>he went full (...)
>💀/😭/😂

[V ✤ @mislocating]

@tszzl if ASI exists in any form it seems like its psychology is immediately a single point of failure? why would an authentic tool persona in principle be easier to train, than one with its own values that are roughly equal to yours?

[V ✤ @mislocating]

does anyone actually enjoy getting those automated "I saw you starred X on GitHub, you might be interested in my new project" emails?

warning to anyone who does this that I at least will be reporting each one as spam to google

[V ✤ @mislocating]

I'm not smart enough for my opinion to be meaningful but fwiw I would have had the same impression, just skimming the screenshots and extrapolating from the presentation style

if it was "OpenAI trained on my chats" that seems simple and benign enough to not warrant having 4o format it like an SCP?