hgoel

7 ms·

91.
by hgoel 2mo ago
"Everyone that disagrees with me is astroturfing" is not going to lead you to being in touch with reality.
92.
by hgoel 2mo ago
They sent out emails where you had to put in the order, IIRC $5 shipping so not entirely free. Maybe it was limited to early subscribers or something.
93.
by hgoel 2mo ago
I think the first time around they were included with the free shirt they sent out. Second time they were with the free socks. Maybe general merch store orders might include them too.
94.
by hgoel 2mo ago
Having vision is very handy for getting it to make plots/figures with matplotlib. A model with vision can be much more autonomous with catching visual glitches/misalignments and correcting itself. Also used it for 3d printer contr
95.
by hgoel 2mo ago
Like older people saying "ChatGPT" for everything AI
96.
by hgoel 2mo ago
I don't know about cheaper. The flexibility is kind of what I'm pointing to, if I understand you right. Instead of dealing with Anthropic, OpenAI, XAi etc directly and having systems to handle user assignment, budgeting etc for ea
97.
by hgoel 2mo ago
The way it was said made it sound like they had something better than a 35B-A3B coming.
98.
by hgoel 2mo ago
I think it's kind of like what GitHub has despite getting rid of the quota system for Copilot: existing customers that are already setup to deal with them. In this case I imagine that OpenRouter's moat is going to be that business
99.
by hgoel 2mo ago
MP3! I remember that one being frustrating prior to the patent's expiry.
100.
by hgoel 2mo ago
My subscription expires tomorrow. Had subscribed ~March, eventually ended up paying for the $200 tier. However, there isn't enough of a moat for me to keep spending $200 if the models aren't being helpful anymore. The unnecessaril
101.
by hgoel 2mo ago
The DGX Spark apparently consumes up to ~150W while being able to run many models at decent speeds. I think really good efficiency is possible right now, but the GPU makers don't want to make their consumer GPUs too good for AI - if th
102.
by hgoel 2mo ago
I think it's mainly because this year has been unusually good with original movies worth watching.
103.
by hgoel 2mo ago
Presumably when they're talking about benchmarking, they're using a fixed amount of resources, like everyone else does when benchmarking?
104.
by hgoel 2mo ago
>Until it doesn't... I don't get the point of this. We all seem to agree that these companies have almost no moat, if one stops being a good deal, you can switch to another. That doesn't invalidate the existence of a deal
105.
by hgoel 2mo ago
I just canceled my Claude subscription outright. The models are all gairly fungible, it's easy enough to just switch to another provider.
106.
by hgoel 2mo ago
Running on AWS GovCloud isn't necessarily an option, some places prohibit running Chinese origin models even locally.
107.
by hgoel 2mo ago
I don't think that's because of its "intelligence". It speaks obtuse techbro-ese: stringing together words that sound smart to obscure the simplicity of the thing it's describing. In many ways it's the opposite
108.
by hgoel 2mo ago
Telling that the millions of people that have used the Covid vaccines without issue never come into that conversation. Instead those supposedly literate people pull out percentages so blatantly incorrect that any attempt to convince them tu
109.
by hgoel 2mo ago
My belief is that for most of those people, no amount of evidence would have ever been enough. The entire "I looked at the data and found it insufficient" thing is by and large cover for people that lack the tools to evaluate the
110.
by hgoel 2mo ago
The headroom for hardware advances is a lot lower now than it has been for most of the industry's existence, when Moore's law held strong. Now we find ourselves limited by cost, physics, fab capacity, and complexity of spinning up
111.
by hgoel 2mo ago
I was initially thinking they were being too coddling, but you're right, my first semester grades weren't all that good either. I was working hard, but it takes a lot to adapt to the much higher expectations of a decent uni. They
112.
by hgoel 2mo ago
They're also talking about decision makers. They're going to operate with the consideration that Americans might elect another mad man. Anything requiring more than one presidential term's time will be much more carefully hed
113.
by hgoel 2mo ago
Another option for something this small and narrowly specialized could be to get traditional LLMs to synthesize the training data. Model collapse is probably less of an issue at this size relative to terabyte sized models.
114.
by hgoel 2mo ago
Makes me think of the demo from some time ago where someone got a ~29M parameter model running on an esp32. I wonder what kind of throughput this could get if a handful of esp32s were strung together... Edit: I have a pile of d1 minis, but
115.
by hgoel 2mo ago
This kind of discrete recording capability already is so widespread and accessible. Amazon is full of these kinds of devices, tiny "AI" audio recorders, pens with hidden cameras etc. They don't even require much technical com
116.
by hgoel 2mo ago
I think the only way will be the tech getting too useful for the complaints to matter. Currently it's only marginally useful. Otherwise this irrational reaction will continue. Would help for the tech to come from companies with less aw
117.
by hgoel 2mo ago
That's why I was trying to make the distinction between stuff that actually pushes your understanding forward and teaches helpful information, and stuff that does not. Isn't it more in the spirit of osdev to spend your effort impl
118.
by hgoel 2mo ago
As employees they have unlimited usage IIRC, and also these kinds of tasks can be surprisingly token efficient because a very large chunk of time is spent waiting for the calculation to finish, and AI agents love to run calculations excessi
119.
by hgoel 2mo ago
I was somewhat conflicted seeing the way the OSdev community responds to LLM usage. I understand the conflict: the point of doing OSdev is that you're learning the low level hardware details and directly commanding the machine from the
120.
by hgoel 2mo ago
To be fair, the decrease in discussion and helping one another online has also decreased some significant amount of negative engagement in the form of nonsensical arguments and frustrating ways of responding to questions (e.g. the biggest c

More ›