When Guinness World Records stopped tracking the record for most beer consumed in one hour in 1989, the standing record belonged to 23-year-old Jack Keyes. He reportedly drank 36 pints in a single hour in Northern Ireland in 1969.
Thirty-six pints in sixty minutes works out to roughly one pint every 100 seconds—a pace that pushes well beyond normal human limits. Given that a standard pint is about 568 ml, Jack Keyes would have consumed over 20 liters of beer in an hour in 1969 Belfast.
Considering the human stomach typically holds about one liter comfortably, the feat is often cited as physiologically extreme. Keyes was only 23 when he reportedly set the record, which remained in the Guinness books for two decades before the category was quietly discontinued in 1989 amid concerns about promoting hazardous drinking. There was no final contest or formal farewell—just the end of the record’s official recognition.
The history of Guinness World Records itself began from a pub debate. In 1951, Sir Hugh Beaver, managing director of the Guinness brewery, argued over which European game bird was the fastest. Unable to find a definitive answer, he realized there was a need for a reliable reference book to settle everyday disputes—an idea that eventually became the foundation for the world records archive.
Open source AI is actually moving at an unhinged pace right now.
I literally hadn't even finished typing up my last Gemma 4 12b benchmark notes before Google went ahead and dropped the official Quantization Aware Training (QAT) checkpoints on Hugging Face.
If you missed the news, QAT basically bakes the compression directly into the training process. Instead of standard post training quantization degrading the model's reasoning capabilities, QAT trains the model with compression in mind.
Unsloth is reporting near original performance at 4-bit with ~72% lower memory footprint. Details in the comments.
Naturally, had to instantly pull the new GGUFs to see what a single RTX 4090 card (24 GB VRAM, Cuda 12.8, ubuntu 22) could do. i fired up llama.cpp engine again
Look at these numbers:
1. Unsloth Gemma 4 26B-A4B IT (QAT Q4_K_XL) flags:
./build/bin/llama-cli -m gemma-4-26B-A4B-it-qat-UD-Q4_K_XL.gguf -cnv -ngl 99 -c 250000 -fa on -v
VRAM Used: 19.5 GB
context: 250,000 tokens
decode throughput: 193 tps
2. Unsloth Gemma 4 31B IT (QAT Q4_K_XL) flags:
Command: ./build/bin/llama-cli -m gemma-4-31B-it-qat-UD-Q4_K_XL.gguf -cnv -ngl 99 -c 60000 -fa on -v
- VRAM Used: 23 GB (Tight, but zero system RAM spillover)
- context: 60,000 tokens
- decode throughput: 47 tps
We are essentially watching hardware bottlenecks evaporate in real time.
An update literally drops before you can finish benchmarking the previous one. What a time to be running local hardware.
If you have a single rtx 3090, rtx 4090, these are the latest gemma models to try this week.
This is the most sober and sobering analysis of AI investing that I have seen. The cannibalizing the passive flows in idices has been buzzing in my head for weeks now...
https://t.co/DNHnWSUUsr
for anyone asking where to learn this stuff:
• RAG → https://t.co/4bzbUIwV5g
• Agentic RAG → https://t.co/IotOiGmV1Y
• AI Agents → https://t.co/nEeMnVJQbk
• Multi-Agent Systems → https://t.co/pavDPVJEFj
• LangGraph → https://t.co/3miEqqFzF0
• LangGraph (code) → https://t.co/v7kxHZXqba
• MCP → https://t.co/lKawRb4etX
• Memory Systems → https://t.co/LSaT2UaPAS
• Evals → https://t.co/vxChxa1kqQ
• Context Engineering → search "Context Engineering Survey" on arXiv
and please skip the "build an ai agent in 10 minutes" videos
build something, watch it fail, then figure out why.
23.5 hours later... there's an app and it's open source.
It tracks activities & sleep. It has full sensor support: HR, SpO2, HRV, Temperature, Motion, etc.
Ken Griffin disagrees with Dario Amodei (Anthropic CEO):
"Data center spending in the United States this year alone is over five hundred billion dollars. You're not going to generate this kind of spend unless you're going to make a promise you're going to profoundly change the world."
So is it hype?
"Of course. How else are you going to get people to write five hundred billion dollars of checks? There needs to be a level of AI is your savior almost."
"In certain areas we know it's going to be profound, whether it's call centers, whether it's helping to improve the productivity of software engineers."
But a Harvard paper gave it a name that stuck.
"There was a recent Harvard paper, they called it AI work slop. It looks good, but if you sort of peel back the onion, the substance isn't there."
Griffin saw it firsthand at Citadel.
"I was with one of my colleagues and he handed me a report we were generating with an AI engine. The first few sentences, wow, that's really insightful. And then you go down below that and it's all garbage."
PS. If you found value in this post make sure to like and repost this tweet + follow @uncover_ai to stay updated with the latest AI news.
See you in the next one:
❗️ Over 30 official Red Hat npm packages were compromised. How they got in:
- A Red Hat employee's GitHub account was compromised.
- Attackers pushed "orphan commits" (detached from branch history) straight in, bypassing code review with no pull request.
- Payload "Miasma" (Mini Shai-Hulud variant) steals GitHub/cloud/Vault/SSH/npm secrets. Rotate everything since June 1.
- The commits added a workflow (ci.yaml) + script (_index.js) that abused npm trusted publishing, requesting a real OIDC token to publish backdoored versions.
Hackers took over high-profile Instagram accounts — including the Obama White House account, a Space Force general, and Sephora — by simply asking Meta's AI support chatbot to change the email address on the target account.
The bot complied.
Meta had rolled out AI support with account recovery powers to all users in March, billing it as "solutions, not just suggestions."
Source: 404 Media