@lmstudio I would love to know the prompt process speed and token generation speed at 1,000 || 5,000 || 10,000 tokens. I've always been a sucker for raw numbers on that.
@arena To me this is a pretty good score. With one exception, the other open models it is behind are large- 800B - 1T+. This one is sliding in at ~450B and ~20B active, so comparatively its doing amazing. Except for DS-4-Flash. That one is punching outside its weight class lol
@logiclpd One thing that's always fascinated me: how in the world do they figure out the right amount of copper? As an outsider who knows nothing of this process, it looks like an arbitrary amount. Like the unit of measurement is "A whole bunch". But I'm sure its actually a precise number
@BasilEsq_@viktorg475@Xianbao_QIAN I don't doubt that they have plenty of ability to lock it down, but at the end of the day the AI company has to find revenue so they have a problem. If the model is too expensive, and the hoops to jump through are too high, users won't use it and corps will find other venues
@otani_ai_memo I can't comprehend how this works without a camera. I thought it was using the gyro or accelerometer: like having it pick a starting point and then move the model with the phone. But then the phone stays static in the second half... and now I'm confused lol. Camera off screen?
@diegocabezas01 IMO: this only affects Anthropic. After getting into a legal fight with the US government, they loudly declared their newest model to be dangerous and practically a weapon, and then were shocked that same government gladly took the given opportunity to slap them down.
@viktorg475@BasilEsq_@Xianbao_QIAN Well then Anthropic and the other AI companies will have to figure out how to get that subsidized, since most users will neither have that kind of money nor the desire to go through intense vetting processes. That includes corporations, which will just outsource the work instead
@Amank1412 I'm sure they'd love to, but they backed themselves into a corner with their marketing. Using "our product is so good that it's dangerous!" as a sales pitch made it too easy for the government to step in and regulate it. Their hands are tied now.
@BasilEsq_@Xianbao_QIAN Anthropic already kicked off KYC via Persona, but users understandably are cautious about giving up 3D scans of their head and face + their government IDs so easily, especially for non-banking purposes, so Anthropic limited the scope of who has to do it to avoid losing too many
@ChShersh If I remember right, this is a holdover from C to avoid overflow. C pretty much takes the stance of "number is number" and uses the cleanest path to it, promoting on arithmetic and making you reconvert.
C# does too, and also makes you do this:
byte c = (byte)(a + b)
@IntCyberDigest I wonder if the folks that had to do Anthropic's identity verification through Persona also lost access to Fable? I'd be pretty upset if I had to give some shady company a 3D scan of my face and didn't even get the upside of being treated as verified
Minimax M3 has toggle for thinking/non-thinking. Im in love. That was one of the things that held me back from using the series more heavily in the past.
https://t.co/YSSxUoGWX3
@Grummz For me, that prompt would be as good as a perm-ban. They already have my phone #, name, address and email. There is no reason to ask me to also disseminate copies of my government ID for the sole purpose of holding it on some server to be stolen one day in a breach.
After wrestling with Qwen3.5's overthinking for a while, I finally gave up and threw workflows at it lol. I haven't had to do that since '24, but man these models are SO good that I can't bring myself to set them aside.
https://t.co/UU1vi7PBXQ
To give you an idea of how tech people name things, my home network + VPN setup includes things like: headscale, a self-hosted DERP, Wazuh, Snort, a llama server and a YARP app.
@cgtwts This is exactly why I will never touch APIs whose billing is simply pay-as-you-go and offers no pre-paid credit option. Yes, you can set limits; but those limits are not always real time. 4 hours can be a LOT of spend. Big companies can handle that; devs & small shops cannot.
If you've never messed with open source LLMs and you jumped on the ClawdBot/OpenClaw hype train: take some time to learn more about how local models work. You likely went through the trouble of getting a Mac Mini, so you now have a nice little test box to play with. Just do it. Turn off Clawdbot/OpenClaw, and make OTHER things with it. Just for a few hours, even.
https://t.co/prScu3rTZ3
Its a crude analogy, but hopefully this helps with some of the folks struggling to understand how to compare the strength of Mixture of Expert models vs Dense models.
https://t.co/vSxt6wUkCZ
This is unfortunate; I enjoyed Discord and it's a shame to see them essentially forcing folks to leave like this.
Realistically- there is no scenario in which someone should upload their government ID for things like chat apps, video games, etc. That is begging for identity theft. All it takes is one of them to have poor PII protection practices, or experience a breach, and you're toast.
This is especially true for companies that JUST made the news for experiencing a loss of customer data, like government IDs. They are the worst ones to give your data to.
Once your ID is out lost to the public, scammers have a LOT of options available to utilize your name and identity for nefarious means. With that ID, and whatever information has been leaked in other breaches like Equifax or bought from data brokers, scammers are as good as you on the internet.
It's unfortunate, but decisions like this effectively force folks off of platforms. Risking your identity and reputation for a messenger app is simply not a good tradeoff. Not when there are alternatives that don't require this.