Readerium

8 ms·

1.
by Readerium 7d ago
DeepSeek v4.1 Flash 31.2% Source: https://huggingface.co/deepseek-ai/DeepSeek-V4.1-Flash#compa...
2.
by Readerium 7d ago
Yeah DeepSeek V4.1 beats this by 15 percent (4 points) on terminal bench 4.0
3.
by Readerium 8d ago
I don't want to be pessimistic, but I can see a single app now showing two different feeds to doomscroll.
4.
by Readerium 9d ago
Question is what does that button do. I bet a lot of lawyers are salivating at this question too.
5.
by Readerium 9d ago
It's not as simple. All our chats are being used by both labs for their future product (unless signed by ZDR). Where should the acknowledgement begin? Who should be acknowledged? The whole world? All the 2B users of AI? If I know perso
6.
by Readerium 9d ago
But then who is the third person in the call.
7.
by Readerium 14d ago
Artificial analysis blog https://artificialanalysis.ai/articles/benchmarking-gpt-6-as...
8.
by Readerium 14d ago
more like 5.7 not 6
9.
by Readerium 14d ago
Its 62 percent when using a neutral harness. https://arcprize.org/blog/astra
10.
by Readerium 14d ago
saturated before (higher degree) AGI-2
11.
by Readerium 14d ago
System Card: https://deploymentsafety.openai.com/gpt-6-astra/gpt-6-astra....
12.
by Readerium 14d ago
Fixed link: https://openai.com/index/gpt-6-astra/ System Card: https://deploymentsafety.openai.com/gpt-6-astra/gpt-6-astra....
13.
by Readerium 14d ago
Working as of 12:30 PM PT https://openai.com/index/gpt-6-astra/
14.
by Readerium 14d ago
https://openai.com/index/gpt-6-astra/ working as of 12:28 PM PT
15.
by Readerium 14d ago
looping sol twice most likely.
16.
by Readerium 15d ago
Washington State University not University of Washington
17.
by Readerium 1mo ago
V4 Pro has vision correct?
18.
by Readerium 1mo ago
Yup same bug I felt.
19.
by Readerium 1mo ago
I expect 5X cache input price hike, and 2X usual input/output hike.
20.
by Readerium 1mo ago
Lol worse than DeepSeek
21.
by Readerium 2mo ago
Nopes, often times especially in math they get paid due to teaching duties (at least in the US). So technically for the math research part they are not getting any stipend.
22.
by Readerium 2mo ago
Exactly, this is an example of "Reward Hacking", that is too common in a lot of cases. Another case I want to highlight is writing GPU kernels as illustrated by the following example: Say I want to generate random number with Norm
23.
by Readerium 3mo ago
3.5 Pro not yet launched, you mean 3.1 pro?
24.
by Readerium 4mo ago
Loll
25.
by Readerium 4mo ago
For MSFT: Just download DeepSeek locally and use it. Or train your own power efficient stack.
26.
by Readerium 5mo ago
AI workloads are all about memory size and bandwidth not compute
27.
by Readerium 5mo ago
LLMs are memory bandwidth bound not compute bound.
28.
by Readerium 5mo ago
Add an iPad mini esque screen on the trackpad of a MacBook. Trackpad of a 16 inch MacBook Pro is humongous anyways. Add a touchscreen display to the trackpad, and give it iPad OS
29.
by Readerium 5mo ago
that is true. gguf does not support any Architecture. for the most recent example, as of April 16, 2026 (today) Turboquant isnt still added to GGUF
30.
by Readerium 5mo ago
perhaps increasing repitition_penalty might be helpful

More ›