goldenarm

6 ms·

91.
by goldenarm 6mo ago
That section misses Cerebras and Groq which are up to 5x faster.
92.
by goldenarm 6mo ago
I prefer that outcome to them acquiring OSS projects and shutting them down once the bubble collapses.
93.
by goldenarm 6mo ago
Has anyone ever seen a person using "context-aware OS" features like Microsoft Recall ?
94.
Ubuntu Linux Will Begin Landing AI Features Throughout the Next Year(phoronix.com)
26 points by goldenarm 6mo ago|23 comments
95.
by goldenarm 6mo ago
gemma4-e4b is 50% better than gemma4-26b in your benchmark, something's wrong
96.
by goldenarm 6mo ago
Rumors and extrapolated from the token price.
97.
by goldenarm 6mo ago
Frontier labs flagship models are ~2T params at the moment, but they intend to ship 10T models like Claude Mythos, which would require substantial datacenter expansion. Same thing for training.
98.
by goldenarm 6mo ago
We may be witnessing a fascinating trend : AI images are making professional-grade imagery look like spam, and natural lighting and blurry images are becoming the new "human" esthetic.
99.
by goldenarm 6mo ago
This article omits important context : these 7 countries have massive hydro power (+geothermal for Iceland) for very little demand. The only countries with <100 g CO2/kWh and >10TWh/y are using nuclear. Large scale batteries
100.
by goldenarm 6mo ago
People have found 0days assisted by LLMs for a while, and none of them wrote hype pieces to find an excuse not to release their 10x bigger model in the middle of a GPU shortage. https://sean.heelan.io/2025/05/22&#x
101.
by goldenarm 6mo ago
Simpler explanation : they don't have enough GPUs to release this much larger model.
102.
by goldenarm 7mo ago
That's a great idea but has anyone benchmarked the performance difference?
103.
by goldenarm 7mo ago
The best move of Apple this decade was to ignore LLMs and let the others burn cash. Now they can use the mature Gemini for $1B. Brilliant.
104.
by goldenarm 7mo ago
Me neither, but roughly a third of my social circle uses AI for news related questions
105.
by goldenarm 7mo ago
Open source AI is mostly Chinese and recites CCP propaganda on many political topics, that's not good either. Thank God we have Mistral and gpt-oss, but they're slowly obsoleted
106.
by goldenarm 7mo ago
How many tokens per second?
107.
by goldenarm 7mo ago
It's more secure, but it would make supply much much worse. Data centers use GPU batching, much higher utilisation rates, and more efficient hardware. It's borderline two order of magnitude more efficient than your desktop.
108.
by goldenarm 7mo ago
Hydro is great for long term storage, but the potential is too small and sparse on the globe to be a scalable solution.
109.
by goldenarm 7mo ago
The definition of renewable is a bit silly, wind and solar come from the sun's nuclear fusion.
110.
by goldenarm 7mo ago
These titles are misleading, they always omit the rest of the grid and the final carbon footprint. In March, the UK emitted 161g CO2/kWh. France did 6 times less CO2/kWh with 2x less renewables ! https://app.electricity
111.
by goldenarm 7mo ago
Google trends show a 15 and 30% decrease that is still going https://trends.google.fr/trends/explore?date=all&q=Reddit,Ne...
112.
by goldenarm 7mo ago
I double checked, the hardware is fine. It's a mix of catastrophic Nvidia drivers and Windows 11 famous lack of optimization. I'm much happier on my 3x cheaper laptop on Linux that can handle three 144hz displays flawlessly on the
113.
by goldenarm 7mo ago
Personal PC. Fresh install from the official ISO with the least bloatware. It's still a nightmare.
114.
by goldenarm 7mo ago
Customer resentment is slow to build up, but once the inertia is visible it's usually too late.
115.
by goldenarm 7mo ago
The specs may be comparable, but not the end result : my $2000 Windows 11 laptop is slower and laggier than the Neo.
116.
by goldenarm 7mo ago
The turing test was passed 2014, before LLMs, and I've never seen a researcher take it seriously.
117.
by goldenarm 7mo ago
AI is poisoning the well of scraping and automation really fast. It generates so much spam, everyone is starting to block robots now.
118.
by goldenarm 7mo ago
Most quality journals are paywalled nowadays, I'm considering to scrape using my cookie, or maybe use archive.is..
119.
by goldenarm 7mo ago
I'm trying to migrate to 100% RSS right now, to avoid the hateful algorithmic editorialization of modern social media. And I'm shocked that almost no paid media provides full articles in RSS anymore, and force me to navigate their
120.
by goldenarm 7mo ago
The lack of backpressure handling nor promise api for postMessage is also quite annoying, I had many OOMs because of it.

More ›