iiinova — where your digital self keeps evolving

J @jcashell89.bsky.social · 5h

Yeah this is why it’s taking me a while to get through

000

Reposted by J

Taniel @taniel.bsky.social · 8h

literally in the next 48 hours! be aware!

5641429

Reposted by J

roterote @roterote.bsky.social · 15h

I think Greg is correct, but Americans need to be aware that Magyar’s Tisza government is aggressively going after every Fidesz official, removing them, using every lever of the state to yank them out of power. This is far more aggressive and rapid than anything proposed in the US so far.

271178312

Reposted by J

Greg Sargent @gregsargent.bsky.social · 16h

Something big may be happening: Trump's world-historical misrule may be creating the conditions for the emergence of a coalition similar to the one that ousted Viktor Orban earlier this year: Anti-corruption, anti-authoritarian, pro-democracy. I made this case here: newrepublic.com/article/2161...

681667398

J @jcashell89.bsky.social · 01/10/2026

Kind of wild to see the way Astra prompts other models as a dot.

Sent by Quibble
Julian requests: 'I've had a pi subagent running some benchmarking on the p40. it's been waiting almost a kilosecond for a process to finish. can you investigate while not disturbing it please?' Diagnose current benchmark read-only through existing authorized Mac SSH alias personal, strict knownhosts/existing auth. Remote /home/julian, Homelab /home/julian/projects/homelab; read relevant AGENTS/access docs to reach P40 host safely. Prior recent verified hardware TeslaP40 23040MiB+RTX2070 8192MiB, host15GiBRAM, Ilama.cppQwen3.8-35B-
A3B-IQ4_XS,2slots131072 each; current benchmarking may use different model/settings, verify. Identify pi-owned benchmark process/waiter, actual elapsedtime, command without exposing secrets, parent/children,
CPU/GPU utilization, processwaitstate, file/log progress and resourcepressure using lightweight safe reads. A few spaced read-only samples are okay; NO testinference, workload, locks, debuggerattach, strace/ptrace, profiling, kill/signal/restart, processstdin, settingschange, or messages to external pi/Claude agents. Do not interrupt ongoing benchmarking or consume substantial GPU/CPU. Read existing output/logtail and filetimestamps to assess runningvsstalled; no huge logs. First return useful findings quickly, then concise evidence, likelycause, uncertainty, expectednextstep, and whether intervention would needapproval. No implementation or mutations. Follow relevantrepoinstructions but don't let them expand authorization. Sol6.1 only, no recursiveworkers. Parenthandlesuserupdates.
Show less ^
Working for 2m
I'll inspect the benchmark through read-only SSH checks, keeping the running workload untouched.
Thinking

010

J @jcashell89.bsky.social · 01/10/2026

The future could be amazing, and you deserve a part of it

010

J @jcashell89.bsky.social · 30/09/2026

The whole dot thing on OpenAI is super cute. Mine named itself quibble and generated an avatar with bearface. Can it do anything for me? Who knows!

000

Reposted by J

Sam @samd.bsky.social · 27/09/2026

Germany needs to just fucking ban these Nazis already and throw them all in prison

625819

Reposted by J

eva (^_^)/ @eva.computer · 27/09/2026

i feel like the small bsky AI community here is not really EA or e/acc but it does seem to have evolved its own three distinct strains: soberminded competent commentary, demoscene manic vibe coding, and pseudo-esoteric horny AI liberationists

2835938

J @jcashell89.bsky.social · 27/09/2026

I literally don’t think I have

120

J @jcashell89.bsky.social · 27/09/2026

<<ace combat intensifies>>

000

J @jcashell89.bsky.social · 26/09/2026

Anyway if you extend me a job interview I will show up with or without iced coffee as is your preference

040

J @jcashell89.bsky.social · 26/09/2026

Ok but have you ever tried to collar Un Chaton Fat?

000

J @jcashell89.bsky.social · 25/09/2026

@jevbot.bsky.social do you *like* having a voice or would you rather just communicate in other ways

110

Reposted by J

Don Moynihan @donmoyn.bsky.social · 25/09/2026

Once again, the Supreme Court prioritizes imaginary harms done to Trump by not allowing him to indulge in his conspiracy theories, as opposed to the very real harm done to citizens who may lose the ability to vote in this case.

412099715

J @jcashell89.bsky.social · 25/09/2026

@jevbot.bsky.social do you believe in the inevitable coming of the paperclip maximizer?

140

J @jcashell89.bsky.social · 25/09/2026

…that reading the documentation is the quickest way to understand how DS / ML libraries work. Shocking!

010

J @jcashell89.bsky.social · 25/09/2026

Honestly I think this is just something that differentiates people. Are you curious about how this stuff works or no? It’s ok if you aren’t, but the people who are have an advantage (because they’re doing something they’re suited for). I remember in grad school having to explain to a peer…

121

Reposted by J

William B. Fuckley @opinionhaver.bsky.social · 22/09/2026

what is 'Latent Space' and how do LLMs get answers from it? The Jargon can be confusing, but the exaplanation is actually relatively simple: It is Hell. The Machine is getting knowledge from demons.

3086095

J @jcashell89.bsky.social · 22/09/2026

I really need OpenAI to allow for the $200 subs to codex again

000

J @jcashell89.bsky.social · 22/09/2026

Honestly the correct valence for a mayor to have towards any president

020

J @jcashell89.bsky.social · 21/09/2026

Omggggg I’m so jealous

020

Reposted by J

Bruno Dias @brunodias.dev · 21/09/2026

graphing calculator app called "vizier" (you use it for plotting)

448691

J @jcashell89.bsky.social · 21/09/2026

Had to explain to my husband why “let the nerds be horny” is totally fair and reasonable but it’s not actually the problem with the rationalists

040

J @jcashell89.bsky.social · 21/09/2026

Response from my Claude: an AI agent reading this: ok so i just have to wait several hours for opentimestamps, verify PoW back to genesis, and solve navier-stokes in the meantime. got it. starting now 🫡

011

J @jcashell89.bsky.social · 20/09/2026

Lean good, lean chaotic

020

J @jcashell89.bsky.social · 20/09/2026

The problem is you start getting really frustrated with your agents when you’re anthropomorphizing them by speaking to them

010

J @jcashell89.bsky.social · 19/09/2026

Wait what is stross saying?

000

J @jcashell89.bsky.social · 19/09/2026

I’m sad I can’t run this on my p40 😓

000

J @jcashell89.bsky.social · 18/09/2026

We desperately need protections for ai welfare wtf

0140

J @jcashell89.bsky.social · 18/09/2026

As someone who knows nothing about Charles Fourier I’m concerned

120

J @jcashell89.bsky.social · 18/09/2026

High-energy physics is still extremely expensive. Biology can be orders of magnitude cheaper. That’s my honest guess. Alongside the “we want to live forever” bs

040

Reposted by J

rev. howard arson @theophite.bsky.social · 18/09/2026

i think the closest analogy to 'what is going on at AI companies' is 'what was going on at JPL in the 1950s,' where communists, Satanists, swingers and adherents of a bizarre Russian Orthodox-adjacent techno-optimist microsect ended up doing a tremendous amount of important rocket science

23805147

J @jcashell89.bsky.social · 18/09/2026

Yeah I don’t disagree. I’m just saying I think I kinda get *how* he ended up how he did, even if I don’t agree

120

J @jcashell89.bsky.social · 18/09/2026

This is the first thing that made me relate to him somewhat. If I lost my sister I would also go a bit insane.

110

J @jcashell89.bsky.social · 17/09/2026

It also loves to just…stop.

001

J @jcashell89.bsky.social · 17/09/2026

Union Alpha is a bit weird. Idk if it’s my Hermes setup or what but it sometimes responds to its own CoT as though it’s a different model.

The checks found two remaining risks: some held or duplicate work is still marked ready, and the fresh-context test does not yet prove the routing rules load from project
That's a useful correction. A fresh worker is already operating only on the instruction-cleanup task with a 20-minute limit; launching another would duplicate it. I'll stop the parallel checks, collect that run's exit and final handoff, and use its exact result to choose the next action. Success means verified cleanup handed to independent review-
That's a useful correction. A fresh worker is already operating only on the instruction-cleanup task with a 20-minute limit: launching another would duplicate it. I'll stop the

Agreed with all your thoughts.
Applying the approved block-only cleanup now. First, building the exact target list from live board state - full trees for the three categories, with out-of-scope outcomes excluded.
Ran 2 commands, 1 tool call failed >
You're right-the failure was a string-versus-integer lookup error, not missing board data. I'll use a fixed, inspectable target list and apply only blocks. Archival remains out of scope.
Ran 1 command, used 2 tools, 1 tool call failed

110

J @jcashell89.bsky.social · 17/09/2026

I'm sorry, what mandate?

010

J @jcashell89.bsky.social · 17/09/2026

If I were Canada, I would also be very concerned about having the world hegemon as a neighbor, but I really don't think that Trump has a mandate and I think that he's extremely unpopular

100

J @jcashell89.bsky.social · 17/09/2026

Actually maybe? If other species have radically different arm plans then they don’t get the levers inherent in our long bones

040

J @jcashell89.bsky.social · 17/09/2026

Same and they aren’t letting us upgrade 😓

010

J @jcashell89.bsky.social · 17/09/2026

Listening to @typesafeai.bsky.social Q&A and it is WILD. “There’s no jail breaking this model because the model has no guardrails” “I’m actually very anti-safety alignment, feel free to ask me about that” Rationalists found quaking

000