@iamgingertrash@yopaonx They are nowhere close to EUV. Not close. They are now designing architectural workarounds that dramatically constrain the entire development cycle. For all purposes, yes - of all human accomplishments EUV is likely the most challenging collection of trade secrets to replicate.
In my work, I have found that conferring agents the ability to edit their own behavior through guidance document is currently a losing enterprise.
Certainly, this will change with further model releases, but right now their decision making is not adequate to keep the junk out of these documents (even Fable). The junk accumulates and the process actually degrades instead of improving.
The current state of self-evolving model behavior is degraded performance and unintended consequences.
Signed up for premium XAI subscription but after a week still don’t have access to Grokbot. No response after a dozen emails.
Instead I got @OpenBot_ powered by Opus 5.5 medium or Sonnet 5.5 medium and I must say - this is awesome! I suspect a lot of the value people found in Grokbot was XAI using Anthropic as the underlying model.
@OpenBot_ is far superior versus Claude desktop and duet for the same use cases (Bot agent and desktop/VM use). The best of all worlds. Excited to see this and other recent projects (T3 code) develop in the open source space.
@DaveShapi I have had sucesss assigning a supervising agent that’s essential role is communicating with opus and a backstop against plain idiotic behavior. Opus still orchestrates the entire task with subagents. That and switching to extra high has largely managed the nerf.
@StockMKTNewz This has absolutely nothing to do with security or geopolitics.
AI chip restriction accomplishes:
1. Massive (albeit transient) limitation on Chinese LLM industry
2. Removed about 1/3 of demand from NVIDIA, cost goes down and availability up
I don’t see any clear indication it will be removed. Quite the opposite - if you notice, most legacy models like o3 and 4.5 have been removed, but 4o remains, assuredly due its fervent supporters. Unlike the others, I do believe 4o can be run relatively cheaply.
I understand that GPT-5 does not meet your use desires, the GPT-5 line is indeed not nearly as intuitive, emotionally resonant, and agreeable as 4o.
I encourage you to also experiment with Claude (after working with personalizing preferences) for your specific case. This model can also be quite emotionally resonant and agreeable, particularly after personalizing the preferences.
SoftBank was the largest shareholder of Nvidia.
but they sold their entire 4.9% stake in Nvidia in 2019 for $3.6B.
It's worth $200B today.
Jensen Huang made fun of Masa in front of him is so funny lmao!
@cairoasmith That is emotionally challenging. What would be the most meaningful technological advancement that would have the greatest impact on the perceived quality of your life?