Does anyone else understand what the difference is between Gemini 3 'Thinking' and 'Pro'? Thinking "Solves complex problems" and Pro "Thinks longer for advanced math & code".
I assume that these are just different reasoning levels for Gemini 3, but I can't even find mention of there being 2 versions anywhere, and the API doesn't even mention the Thinking-Pro dichotomy.
- "Thinking" is Gemini 3 Flash with higher "thinking_level"
- Prop is Gemini 3 Pro. It doesn't mention "thinking_level" but I assume it is set to high-ish.
Really stupid question: How is Gemini-like 'thinking' separate from artificial general intelligence (AGI)?
When I ask Gemini 3 Flash this question, the answer is vague but agency comes up a lot. Gemini thinking is always triggered by a query.
This seems like a higher-level programming issue to me. Turn it into a loop. Keep the context. Those two things make it costly for sure. But does it make it an AGI? Surely Google has tried this?
I don't think we'll get genuine AGI without long-term memory, specifically in the form of weight adjustment rather than just LoRAs or longer and longer contexts. When the model gets something wrong and we tell it "That's wrong, here's the right answer," it needs to remember that.
Which obviously opens up a can of worms regarding who should have authority to supply the "right answer," but still... lacking the core capability, AGI isn't something we can talk about yet.
LLMs will be a part of AGI, I'm sure, but they are insufficient to get us there on their own. A big step forward but probably far from the last.
> When the model gets something wrong and we tell it "That's wrong, here's the right answer," it needs to remember that.
Problem is that when we realize how to do this, we will have each copy of the original model diverge in wildly unexpected ways. Like we have 8 billion different people in this world, we'll have 16 gazillion different AIs. And all of them interacting with each other and remembering all those interactions. This world scares me greatly.
Advanced reasoning LLM's simulate many parts of AGI and feel really smart, but fall short in many critical ways.
- An AGI wouldn't hallucinate, it would be consistent, reliable and aware of its own limitations
- An AGI wouldn't need extensive re-training, human reinforced training, model updates. It would be capable of true self-learning / self-training in real time.
- An AGI would demonstrate real genuine understanding and mental modeling, not pattern matching over correlations
- It would demonstrate agency and motivation, not be purely reactive to prompting
- It would have persistent integrated memory. LLM's are stateless and driven by the current context.
- It should even demonstrate consciousness.
And more. I agree that what've we've designed is truly impressive and simulates intelligence at a really high level. But true AGI is far more advanced.
Humans can fail at some of these qualifications, often without guile:
- being consistent and knowing their limitations
- people do not universally demonstrate effective understanding and mental modeling.
I don't believe the "consciousness" qualification is at all appropriate, as I would argue that it is a projection of the human machine's experience onto an entirely different machine with a substantially different existential topology -- relationship to time and sensorium. I don't think artificial general intelligence is a binary label which is applied if a machine rigidly simulates human agency, memory, and sensing.
This looks awesome! Is there a way to log in some way that doesn't leave your credentials saved in a text file or your bash history? I would use this at work, but I think my sysadmin would eat me alive if I was storing the login credentials in plaintext
I am going to implement the option to leave the password blank, so you'll get prompted for a password on connection every time, so nothing will get stored.
In the next release I am going to use Keyring to store credentials on the operating system's credential store. Stay tuned :)
Just finished college, so I'm not too long in the tooth. I remember reading your writeup on the development of Galak-Z, and liking it so much that I bought it. I'm pretty sure it was one of the first steam games I ever purchased. I didn't end up liking the game all that much, but I definitely continued to enjoy your writing!
SVG's are so fun! The amount of logic you can put into an SVG is absolutely wild given what they're generally used for. My desktop wallpaper uses embedded JS to create a random Barnsley fern fractal every time the system starts, so it's different every time. The only problem is that only browsers have the full SVG spec support I needed to get it to work correctly, but that's nothing that systemd and puppeteer can't fix.
I feel like if I mention technology X in my system context for Gemini, there is a 100% chance that when I ask for hiking recommendations Gemini will say "As a user of technology X, you would appreciate the beauty and elegance of the Cuyamaca National Forest"
My introduction to JPEG-XL was by 2kliksphillip on YouTube, he has a few really good analyses on this topic, including this video: https://youtu.be/FlWjf8asI4Y
This article (and it's second part) couldn't have come at a more helpful time. I have such an unhealthy relationship with caffeine (and sugary drinks in general), that I would literally go days without drinking any water. If it didn't come from a 7-11 or a coffee machine, I wasn't going to drink it.
Just this week, I decided to quit cold turkey, and drink nothing but water, and it has been eye opening. I didn't realize how much I depended on the drinks to get me through the day. But going just-water is definitely the only way I could have done this. Something about knowing I have nothing to look forward to besides water helps me just drink water, instead of waiting until I can find some energy drinks, and getting dehydrated in the process.
Not everyone is going to have as unhealthy relationship with caffeine as me, but I'm hoping my kidneys and gut will thank me. I know 22 year olds seldom suffer from heart attacks, but I'm not going to be 22 forever, so might as well start now.
I would literally go days without drinking any water.
The idea of that is horrifying. I can't imagine voluntarily putting myself through that in the absence of some sort of extreme drought situation. Personally, I'd be miserable without 64 - 128oz of water first thing in the morning. I've long suspected that a lot of people who think they're addicted to caffeine are really just chronically dehydrated.
I think there's a lot of truth in that. The most surprising thing to me has been how much water I've been drinking. Around 3 quarters of a gallon a day, which is far more than I would drink of anything before. And I think that will have a whole bunch of benefits beyond just the lack of caffeine.
I think I was less caffeine addicted, and more sugary-caffeinated drinks addicted.
I used to not drink coffee, joined the high-end VFX industry, started to drink a lot, e.g., 3 big dark mugs at work each day. We had top-of-the-line espresso machines and great coffee so it was easy. Something I noticed is that on weekends, I had systematically huge migraines, those were soothed by drinking more coffee...
The issue is that caffeine has a vaso-constricting effect on blood vessels [1], which then, when they re-expand by lack of caffeine creates the migraines.
One day, I decided to stop, I had an horrible week, constant migraines, it eventually stopped and I never really had coffee since then. Only 2 times during conditions where I needed which added the benefit that the caffeine had a really strong effect then.
I'm certainly missing the taste and ritual, so I replaced coffee with green tea, not as good but no migraines!
personally I've found the best way to go cold turkey is when you have a cold or the flu.
if you're going to feel like shit you might as well have something to blame it on and once you finally get over it you'll feel good enough to not worry about the lingering withdrawals
I'm very sensitive to caffeine and get jittery and anxious around the 80mg mark. But I also adore coffee and keep ramping up. Eventually I'm at 240mg per day and can't actually function, I'm just a terrible person that doesn't sleep much at night. I really wish we'd make caffeine harder to get :D
Speaking of which, time to cycle down. Back to tea or water.
I've been using qtile as a replacement for hyprland for a few months now, and I don't have any complaints at all. It's an extremely functional piece of software, and at least for me it seems to be more stable than hyprland was.
It's so out of the way isn't it? I have to switch to using a Mac for work and I can't get the Aerospace filing window manager to work like Qtile does. It's a shame.
On that note, is there any good window management paradigm on MacOS? Tiling just doesn't seem to work well tacked on compared to full-fledged i3, and MacOS' seperation of applications from their windows makes multi-tasking a giant pain. Windows is miles ahead in this regard, and Linux of course offers options for every paradigm imaginable, so it leaves MacOS feeling particularly mediocre.
Is it? I've found Raycast to be one of the greatest boosts to my productivity of any app on any platform. Then again, I was doing fine with Rofi + i3 before.
We weren't given a chance to see what was being claimed and properly respond to it. Our response at the end of the article was to this prompt, which was in the first and only email we received, in English:
> I am preparing an article on the use of your secure personal data phone solution by drug traffickers and other criminals. Have you ever been contacted by the police?
The claims in the main story strongly indicate they're not talking about GrapheneOS itself but rather companies selling closed source forks of it with significant modifications. They refer to features which don't exist in GrapheneOS. Supposedly GrapheneOS which is freely available from https://grapheneos.org/install/web and https://grapheneos.org/releases with sources on GitHub is distributed on the "dark web" and promoted via unlisted YouTube videos. They're clearly conflating products which market themselves by saying they're using GrapheneOS with the upstream project those are forked from. These are largely sketchy products and we regularly have to deal with them infringing on our copyright and trademarks.
One of these companies marketing products claiming to use GrapheneOS, ANOM, turned out to be a company run by the FBI as a sting operation which was hiring criminals to sell phones to other criminals. ANOM told people what they were getting was GrapheneOS when it was actually a mix of GrapheneOS and LineageOS code. The FBI was broadly facilitating crime in Europe by providing them devices they considered secure and safe to use while disregarding most of it to avoid exposing their operation. They were also misusing our brand and harming our reputation us through this. A lot of the claimed criminal usage was directly engineered by the FBI. A detailed podcast episode on this:
It says that if we don't cooperate, they'll take similar actions against us they did against 2 named secure phone companies. Those actions were taking over their servers and criminal charges. It's clear what they want is a backdoor to have access to devices they're unable to exploit due to the advanced exploit protections. They're threatening that if this is not provided, they'll go after us as they did companies they said were collaborating with criminals. They likely consider providing freely available open source software which anyone can use for any purpose to be collaborating with criminals.
The main result will be OVH losing our business to a Toronto colocation provider for important non-static content (discussion forum, email, Matrix, Mastodon, attestation service), Vultr (American) for our anycast DNS + exotic webserver locations, Netcup (German) and perhaps another 1-2 companies for NA/EU web servers where Vultr is extremely overpriced due to double the costs for the same specs and metered bandwidth (it's great for exotic locations and BGP support for our anycast though).
There's another article here, but the paywall isn't bypassed by archive sites (we've read it though):
I assume that these are just different reasoning levels for Gemini 3, but I can't even find mention of there being 2 versions anywhere, and the API doesn't even mention the Thinking-Pro dichotomy.
reply