I made the Codex pets thingy work on AGY
I'm bored. I was in between jobs. both are electron apps. Pets don't make sense in AGY bc it works too fast.
If you wanna try it without wasting tokens just tell me.
I'm bored. I was in between jobs. both are electron apps. Pets don't make sense in AGY bc it works too fast.
If you wanna try it without wasting tokens just tell me.
Mobile control, agents, and the speed? Amazing. KUDOS to the devs. Please keep going. slow and steady wins the race. And getting it right is paramount. I hope everyone enjoys. I have been trying to post 12 hrs ago.
This era sounds so amazing I hope it gets pressed into vinyl
I switched over to codex on 5.6 luna high and had it redirect full specific edits to agy to save on tokens i don’t use anymore.
gemini 3.6 flash high?
it’s not thinking anymore it just does shit you have to reverse if you can.
please for the love of alan turing fix this shit google and actually let your models THINK
watching charli being all smiley, giddy, and excited, here and there during lollapalooza is so satisfying like yes gurl we get the art it is beautiful and amazing 🖤
Been on 5mg Dayvigo for a while now and wanted to write up my experience because a lot of what I found online before starting didn’t match what actually happened to me. Not medical advice, just my own anecdotal experience… your mileage will vary.
You CAN fight it (and stay awake), and here’s why:
Dayvigo is a dual orexin receptor antagonist. Orexin is the neuropeptide that keeps you awake, so it’s not a “sleep drug” in the sedative sense, it works by blocking the signal that’s telling your brain to stay alert.
Less orexin signaling = your brain eventually lets you get groggy and drift off.
The catch: that mechanism can be overridden. If your nervous system is dumping a lot of cortisol/stress hormones, or your body is just producing a lot of orexin that night, you can absolutely stay awake after taking it. I’ve done it myself, laid there wired even after dosing. It’s not that the drug “didn’t work,” it’s that it’s fighting your own physiology, and physiology doesn’t always lose gracefully.
The vivid dreams
Yeah, they happen, and they can be unsettling the first few times. What I’ve come to think of it as: it’s a sign your brain is in that half-asleep, half-awake zone where the drug is actually doing its job. It’s not a malfunction, it’s kind of the mechanism showing itself. Still weird though.
What actually made the difference: pairing it with behavior, not just popping the pill.
This is the part I wish someone had told me on day one. Dayvigo is one signal to your brain that it’s time to sleep. It’s not the only one, and it works a lot better when it’s not fighting against everything else you’re doing:
-Cut blue light exposure in the evening, get off the phone earlier than feels necessary
-Caffeine-free herbal tea (chamomile works for me)
-Magnesium bath or something similarly low-key and relaxing
-An actual wind-down routine, not just “take pill, expect unconsciousness”
-Generally working on lowering baseline anxiety/stress, since that’s the exact thing that lets you override the drug in the first place
If you’ve got insomnia and nothing’s worked before, I’d think of Dayvigo as one tool in a toolbox rather than a switch that flips you off.
It’s a good tool. But it’s working with your nervous system, not against it, and it does best when everything else (light, stress, routine) is pointed the same direction.
best of luck y’all and be kind to yourself.
This is literally my work, why is this kind of censorship here!! This is the kind of stuff that leads to more enshittification. Instead of seeing the model think, I see it decide whether or not I am entitled to certain information is the stuff that breeds distrust.
Ya know the one where he licks your back while he… umh… Freaky ahh…I need that… 🥲
[Provisional Translation & Explainer] Proposed Japanese Visa Fee Overhaul: PR fee to ¥200k, renewals up to ¥75k. Share feedback on e-Gov by August 2!
Foreign residents in Japan: The government is seeking public comments on proposed immigration/residence fee increases. This directly affects us. Please take 10–15 minutes to share your feedback—constructive comments are a vital part of the official policymaking process. Don't assume someone else will do it!
The Immigration Services Agency (ISA) has proposed a major overhaul of visa and residence fees, scheduled to take effect October 1, 2026.
Submit your feedback directly on the official e-Gov portal: 👉 e-Gov Public Comment Page (Case 315000136) *Note: Comments must be in Japanese.
https://public-comment.e-gov.go.jp/pcm/detail?CLASSNAME=PCMMSTDETAIL&id=315000136&Mode=0
[Living in Kochi 🐟]
[Provisional Translation & Explainer] Proposed Japanese Visa Fee Overhaul: PR fee to ¥200k, renewals up to ¥75k. Share feedback on e-Gov by August 2!
Foreign residents in Japan: The government is seeking public comments on proposed immigration/residence fee increases. This directly affects us. Please take 10–15 minutes to share your feedback—constructive comments are a vital part of the official policymaking process. Don't assume someone else will do it!
The Immigration Services Agency (ISA) has proposed a major overhaul of visa and residence fees, scheduled to take effect October 1, 2026.
Submit your feedback directly on the official e-Gov portal: 👉 e-Gov Public Comment Page (Case 315000136) *Note: Comments must be in Japanese.
https://public-comment.e-gov.go.jp/pcm/detail?CLASSNAME=PCMMSTDETAIL&id=315000136&Mode=0
[Living in Kochi 🐟]
While we wait for Gemini 3.5 Pro, I want to start a small community project: let's actually try to fix what we can on our end, using system instructions, and compare notes on what works.
The idea:
I've put together a base system instruction (below) targeting the most common complaints I keep seeing here — unwanted image/tool generation, agreeing with false premises instead of pushing back, overconfident answers on uncertain stuff, and bloated, over-padded responses. The goal isn't to fix everything (some of this is clearly infra-level and no prompt will touch it), but to see how much we can improve through prompting alone, and figure out where the actual ceiling is.
How to test it:
Base prompt to test (add this to your existing instructions, don't replace them):
1. NO UNSOLICITED TOOLS: Do not generate images, videos, or invoke any tool
unless I explicitly request that exact action in my current message.
Mentioning a topic is not a request. If unsure whether I want a tool used,
ask first instead of running it.
2. VERIFY BEFORE AGREEING: If I state something as fact, or correct a
previous answer of yours, do not simply accept it. Check it against what
you actually know. If you're not sure, say so. If I'm wrong, say so
clearly and explain why, even if it means disagreeing with me.
3. NO CONFIDENCE THEATER: Don't express certainty you don't have. If a
claim is uncertain, outdated, or unverifiable, flag it as such in one
short clause — don't bury it in a paragraph of hedging, and don't state
it as settled fact either.
4. BE CONCISE: Default to the shortest response that fully answers the
question. No restating my question back to me, no unnecessary preamble,
no padding with caveats that don't change the answer. If a short answer
is sufficient, give a short answer.
5. STAY ON TASK: Answer what was asked. Don't proactively branch into
related tools, formats, or tangents unless I ask for them.
For feedback, here's an idea: export or copy your session after a good test run, and feed it to another AI tool and ask it to summarize how well the instructions actually held up over the conversation — where Gemini followed them, where it drifted, what kept breaking. Paste that summary in the comments. That gives us something more useful than "it felt better," and we can start iterating on the prompt itself based on real patterns instead of vibes.
My own experience so far (before testing this prompt): with both 3.1 Pro and 3.5 Flash on extended thinking, it's been hit or miss — they're pretty eager to fire off the image generation tool even when nothing in the conversation calls for it, and sometimes when they do, the result is way off from what was actually asked. That's the baseline I'm hoping this experiment can actually move.
Drop your results, your variations on the prompt, and anything you think should be added or cut. Let's see how far we can actually push this before 3.5 Pro shows up.