SYS 64738 in progress. Coding in 1984 forever.
nusquamu
Joined May 2018
- Tweets755
- Following253
- Followers33
- Likes119
trashsimmonds retweeted
Hey Radelaide folk - a kindly redditor has put together a bunch of resources for the upcoming aurora:
reddit.com/r/Adelaide/commen…
trashsimmonds retweeted
What's the last course you took? Hopefully in the past year. could be free or paid.
trashsimmonds retweeted
We topped Retention (Satisfaction) in Front-end Frameworks in the State of JS survey for the 3rd year in a row!
Thank you everyone who participated. It is great to hear how happy you are using @solid_js to build your projects.
trashsimmonds retweeted
Received maybe the most 'german' compliment ever yesterday:
I think you are doing the wrong thing, but you are doing it very well.
trashsimmonds retweeted
After 2 years of research, two beta releases and countless hours of R&D, the @solid_js Team is proud to announce the release of SolidStart 1.0! 🚀🎉
SolidStart has already made an impact on the broader JS community. We’re excited to see what’s next.
solidjs.com/blog/solid-start…
trashsimmonds retweeted
We have just released a new, free, fully Open-Source LLM. It is also the best Open-Source model by a long way on the Arena-Hard benchmark today.
Also worth pointing out that MT-Bench is really capped out at this stage - I think the community should move on to a new benchmark ...
Smaug - the best open-source model in the world, rivals GPT-4 Turbo.
Llama-3 70b was the best OS model till today. Today, we are happy to drop a significantly better model, Smaug-Lllama-3-Instruct.
First, with Smaug, we see a significant improvement in the MT-bench. This score is correlated with human eval.
1st turn
smaug-70b 9.4
llama3-70b 9.2
gpt-4-turbo 9.37
2nd turn
smaug-70b 9.0
llama3-70b 8..8
gpt-4-turbo 9.0
Average
smaug-70b 9.2
llama3-70b 9.0
gpt-4-turbo 9.18
My big knock on MT-bench and human eval is that they only address simple prompts that humans come up with.
In the real world, agentic tasks require complex reasoning and planning. Arena Hard is a new benchmark that measures an LLM's ability to solve complex tasks.
On this benchmark, Smaug makes significant gains over Llama-3 and scores 56.7 over Llama-3's score of 41.1
Smaug is currently the best open-source model globally and rivals GPT-4 Turbo. (Hf link in alt)
Open-Source Is Far from Dead 🚀🚀🚀🚀🚀
trashsimmonds retweeted
I'm not sure anyone else is picking up on the magnitude of this development.
What we all consider the "free encyclopedia of human knowledge", is no longer accessible unless you're using the tools they specify are deemed ok to view them with.
trashsimmonds retweeted
O/T - anyone else experiencing this inability to access Wikipedia unless using an approved browser/internet provider?
As a coal-face canary dev this troubles me, and paints a dismal picture of the future of interwebs access.
trashsimmonds retweeted
AIML has achieved Facepalmbook user level of intelligence and mental gymnastics:
"Question: Sally (a girl) has 3 brothers. Each brother has 2 sisters. How many sisters does Sally have?"
reddit.com/r/LocalLLaMA/comm…
trashsimmonds retweeted
The one positive trait in the near/medium-future is leveraging AIML for assisted scholastic stuff, changing the rule books so people can actually learn something useful, rather than recite things to pass arbitrary puzzles to get "grades".
trashsimmonds retweeted
I mean, this is a better more contextual result with citations (I checked them) than a well crafted question on Reddit et al. @YouSearchEngine
