Iolaum

6 ms·

61.
by Iolaum 3mo ago
TSLA and SPCX issues are not short term.
62.
by Iolaum 3mo ago
Likewise here, my setup costs 2k+ more eur as well. Given that memory bandwidth remains the same I don't think it's worth buying if you have the 1st gen framework desktop. Usecase caveats apply.
63.
by Iolaum 3mo ago
Model Looks amazing! Even more important, subjectively, is that this model will run very well on Strix Halo (e.g. Framework Desktop), DGX Spark kinds of devices. Looking forward to Unsloth dynamic mtp quants. P.S. Looking at the HF release
64.
by Iolaum 3mo ago
That's only one factor of what people are worried about.
65.
by Iolaum 3mo ago
What if a US Company takes a Chinese AI open model, makes some minimal changes in it (e.g. when you ask it who are you :p ) and then hosts it as their own open model. Without any financial transaction (due to entity list restrictions). Will
66.
by Iolaum 3mo ago
IMO the biggest argument against "sharing" your code with LLM providers is that your approach (on a high level) will be available to your competitors on the next model release assuming they ask the right questions. Not sure how mu
67.
by Iolaum 3mo ago
Doubt they will do it as long as Anthropic is leading in business adoption. If they become the top dog with a good lead, all bets are off. Hopefully by the time open models will be even better than gpt-5.6 sol xD.
68.
by Iolaum 4mo ago
LLM's as a technology - currently - are stateless. The memory layer is controlled by the agent. I m surprised with the gpt-6 reference unless it has to do with vertical integration between the agent and the layer.
69.
by Iolaum 4mo ago
The name is of Greek origin https://en.wikipedia.org/wiki/Akritai from frontier soldiers guarding the Byzantine empire's borders.
70.
by Iolaum 4mo ago
At full quantization GLM 5.2 may be close to GPT 5.4. But at Q2 or whatever one needs in order to run it on a pro-sumer device it will be worse. Also I m not sure where you are getting the under 2k value. I bought a Framework desktop 128GB
71.
by Iolaum 4mo ago
Two big names left GDM recently. Could be a coincidence, but where's the fun in that? :p
72.
by Iolaum 4mo ago
Why unquantized instead of Q8 ?
73.
by Iolaum 4mo ago
LLM's learn by training on examples more than by training on documentation. Especially since examples are usually bigger in data size.
74.
by Iolaum 4mo ago
Haven't used for actual coding but was testing locally - for example running some swebench instances - whether qwen-3.6-35b-a3b@Q8 was better than qwen-3.5-122b-a10b@Q4. With MTP the former runs at around 55t/s and the latter at a
75.
by Iolaum 4mo ago
it's a feature, not a bug
76.
by Iolaum 4mo ago
Unfortunately Mistral is not close to the frontier. Their last release Mistral Medium 3.5 128B is near the performance[0] of QWEN-3.6-27B, a much smaller model that was released earlier. It's good that they exist, and I hope they catch
77.
by Iolaum 4mo ago
> The world is a bit bigger than US and China With respect to AI capabilities is it really? I don't see anyone else producing frontier closed source LLM's or frontier open source LLM's outside of US and China.
78.
by Iolaum 4mo ago
Does this mean Telsa would get out of the S&P index, or that SpaceX would come in? Hope for the former.
79.
by Iolaum 4mo ago
They (Anthropic) don't need to "look" at the data. Just use them to train the next model and then their competitors to ask the new model how they can improve their product :p
80.
by Iolaum 4mo ago
One popular solution lately has been instead of banning too much, because of the danger of false positives, to use vouch [0]. Trusted people get vouched and you prioritize their actions. Unknown people (or agents) need to gain trust to be v
81.
by Iolaum 4mo ago
My physics bias would like to see earth forming while it's constituents were orbiting around the sun. In any case, nice visualization.
82.
by Iolaum 4mo ago
Why not ask Claude design to write a document fully specifying the prototype?
83.
by Iolaum 5mo ago
Good, no fast track to the S&P 500 !
84.
by Iolaum 5mo ago
Strix Halo chips have around 210+ GB/sec gpu memory bandwidth and announcements put the new nvidia chip at around 300GB/sec gpu memory bandwidth. I 'd say that is an improvement if you want to run local llm inference. Still w
85.
by Iolaum 5mo ago
The LLM's had to learn that from somewhere :p
86.
by Iolaum 5mo ago
I 'd say he is an outstanding engineer as well. He may favor output over security more than outstanding engineers at 2025 but in the 2026 world what he does is impressive. And with OpenAI's resources he has turned OpenClaw's
87.
by Iolaum 6mo ago
Many people noticed the math doesn't add up, e.g. in the CEO's interview here [0]. Am wondering if they are betting on the stock going up because of the news so they would then be able to fund the deal. [0]: https://www
88.
by Iolaum 6mo ago
We shouldn't need a lawsuit. The legislative branch should pass a law clarifying those things, that's their job.
89.
by Iolaum 6mo ago
$200 for you is not the same as $200 for them
90.
by Iolaum 6mo ago
I don't know their strategy but I wish they would double up on the open source ecosystem by sharing their innovations like Chinese labs do and use the ones shared as well. I think models in the range of 50B - 250B still have a lot of r

More ›