Harald Klinke
banner
harald-klinke.de
Harald Klinke
@harald-klinke.de
Digital Transformation, Data Science, AI, Munich
🙋 https://www.harald-klinke.de/
📔 https://link.springer.com/book/9783031881305
Meta, Amazon, and Google are expanding their global AI infrastructure beneath the oceans:
Project Waterworth (Meta): 50,000 km across five continents
Fastnet (Amazon): Maryland ↔ Ireland, 320 Tbps
Sol (Google): USA ↔ Bermuda ↔ Azores ↔ Spain

#CloudComputing
www.cnbc.com/2025/11/08/b...
Underwater cables are a vital piece of the AI buildout and internet — investment is booming
Over 95% of international data and voice traffic travels through subsea telecom cables. Tech giants Meta, Google, Amazon and Microsoft are investing for AI.
www.cnbc.com
November 10, 2025 at 6:50 AM
Thanks again, Mario, especially for pointing to sitegeist.ai. That’s exactly the missing link I was curious about. Really appreciate you sharing the details!
Sitegeist - Your AI Companion for the Web
sitegeist.ai
November 7, 2025 at 9:11 PM
Reposted by Harald Klinke
Ahhh, I just realized you mean the paper generator above.

I built a browser agent: sitegeist.ai

I then pointed it at:

- YouToube videos (extracted transcripts)
- BlueSky threads (extracted text, img urls for charts and their alt text)
- Google Scholar for references that fit
Sitegeist - Your AI Companion for the Web
sitegeist.ai
November 7, 2025 at 8:46 PM
Reposted by Harald Klinke
I'm happy to present my latest oeuvre: a "scientific" paper on the use of LLMs by newspapers, generated by an LLM, based on @schenior.bsky.social and my BlueSky threads and YouTube video transcripts over the past few days.

This would pass peer review of some C/D confs/journals I was a reviewer on..
November 7, 2025 at 8:53 PM
Thanks, really impressive work. Appreciate the insight!
November 7, 2025 at 9:05 PM
Fascinating experiment, would you mind sharing a bit more about the actual tool chain behind it?
November 7, 2025 at 8:29 PM
While Microsoft struggles with power shortages for its AI data centers, Google looks beyond Earth.
With Project “Suncatcher,” the company plans to deploy solar-powered data centers in orbit by 2027 — operating continuously in sunlight.

#AIInfrastructure #Energy #SpaceTech
Google wants to build solar-powered data centers — in space
The tech giant released a paper Tuesday outlining plans to launch its AI chips into low-earth orbit, allowing them to run off solar power.
www.semafor.com
November 5, 2025 at 8:18 AM
Reposted by Harald Klinke
As for the METR paper, it's definitely needed research, because claims by AI peddlers are severly overblown imho.

However, I do think that LLM-assisted development can be a huge net benefit. If you put in the time to establish your own workflows and understand when and when not to use an LLM.
July 16, 2025 at 6:23 PM
Scaling Reinforcement Learning (RL) is emerging as the next frontier after pretraining.
“RL compute for LLMs has scaled massively, but our understanding hasn’t kept pace.”
#LLM #AIResearch #Scaling

bsky.app/profile/nato...
New post recapping the biggest paper in reasoning/RL this year since the DeepSeek R1 report. It does a great job highlighting how RL (and post training really) is far more of an art than a science (pretraining is a hard science).
www.interconnects.ai/p/the-new-rl...
The new RL scaling laws
The most covetable research.
www.interconnects.ai
October 20, 2025 at 7:01 PM