Hani Safadi

safadi makes streaming #fun #uga

#uga today safadi teaching about streaming

teaching about streaming #uga and safadi is the best instructor!

@wesmckinn Curious to know if you used any genAI coding assistance while developing it.

I've used Gmail for 20 years. Almost 2M emails, 150K attachments. Rather than let Google hold my data hostage, I built msgvault: local-first email archive with a terminal UI and MCP server, powered by DuckDB. Open source, single Go binary. wesmckinney.com/blog/announcin…

I had a great time sharing thoughts on teaching AI for analytics today @TerryCollege linkedin.com/feed/update/ur… #misatuga

@AOM_CTO May be?

Do we still have active community here?🤔

#uga learning about #dataengineering

#uga learning about data streaming #terry

terry students are learning a lot @hanisaf

#mis terry is great

demonstrating stream processing in the terry college #mis

@emollick A case for machine behavior dx.doi.org/10.1038/s41586…

It is so strange that we have to figure out how (or even whether) our latest software does critical functions that would normally have to be carefully designed. More like biology or psychology than computer science.

More chatter and less connection. Recent research from the @UniversityofGA into bots on X and Reddit show they boost individual user engagement but curtail connections between people. Learn More: terry.uga.edu/more-chatter-a…

@tdinh_me is it possible to see thinking tokens for o1?

Deepseek R1 is so cool I had to build a custom parser for it. TypingMind now can show "thinking" text in its own collapsible block.

@OpenAI 3 below are the results of 4o, o1-mini and o1-preview. 4o and o1-mini are both bad prosody (ok content). o1-preview is the only one with a good start for a poet to edit. If you are Arabic speaker, feel free to comment, agree or disagree! end of 🧵

2 So far all LLMs flunked this task. The results are typically good content but messed up prosody. Trying @OpenAI o1-preview today gave me for the first time an output with a 80-90% correct prosody. What is interesting is that the thought process was in English 🤯

🧵1 One test I use for LLMs is to write classical Arabic poetry about a certain topic. It's a hard task given that the choice of words should not only meaningfully reflect the chosen topic but also comply with a complex set of constraints (en.wikipedia.org/wiki/Arabic_pr…)

@tdinh_me While the visualizations are cool, the JS-based agent cannot analyze real-world data, unlike ChatGPT Python-based code interpreter, which's why I'm keeping my pro membership. I understand you want to run everything on the client side, but have you considered Python @pyscript_dev?

This is the best LLM UI to use with your API key:

@emollick The NN provides a different implementation of the game engine, based on input/output pairing rather than rule-based.

Wow, diffusion models (used in AI image generation) are also game engines - a type of world simulation. By predicting the next frame of the classic shooter DOOM, you get a playable game at 20 fps without any underlying real game engine. This video is from the diffusion model.

On Aug. 7, the U.S. District Court for D.C. ruled Google was guilty of creating & maintaining an illegal monopoly over internet search. UGA professor Hani Safadi answers a few questions about knowledge monopolies & why they stifle innovation: terry.uga.edu/knowledge-mono…