TrendPulse

Preferably up to date with modern AI

source: https://x.com/blelbach/status/2073232846731301347?s=20

forward to the day when good-looking humanoid robots live among us

Hi everyone, I am a computer scientist by passion and profession, and I wanted to offer some food for thought on what, data in hand, risks becoming the greatest and most concrete danger of current technological evolution. Progress is running faster than we can truly understand it, so I hope this can help provide a better grasp of the current situation. I have structured the problem into 5 precise steps: The infrastructure monopoly : Developing the most advanced AIs requires an amount of m...

The more I use AI, the more one thing about its economics bothers me. What unsettles me is how the economics seem to work. These models cost enormous amounts of money to train and run, yet companies keep making them cheaper or even free for users. Investors are pouring in billions with the expectation that profits will come later, but it's hard to see where those profits actually come from if AI keeps driving down the value of the work it's replacing. What am I missing? submitted by ...

I have been thinking about the Fable 5 to GLM-5.2 sequence as one event rather than two. June 9, Anthropic ships Fable 5, the Mythos line opens to the public for the first time, SWE-bench Verified at 95 percent, people calling it the best model they had ever used. June 12, US Commerce sends the export control letter, by night the model is gone globally, including for US citizens, because Anthropic could not surgically restrict only foreign nationals. Ninety six hours end to end. Four days af...

after the massive improvement in coding models I feel like creative models have regressed and become way more corporate sounding. When I try to ask for ideas or even simple titles for things I find the lists very unimaginative and rarely something I havent already considered. creative writing and plot type questions are also super pedestrian and dull. considering these models have been trained on so much literature I rarely get surprised by any ideas coming from a model. Was it always like th...

I have two colleagues that always consult claude on like every step of what they are doing. One always copy paste AI reply, so you can imagine any time on chat, he always talk in “points”. The other one, every time we did something he wants to double check with AI even like the smallest update on non-critical thing, like he always “let me check with claude”. Don’t get me wrong, i do consult AI and do vibe code (although I have decent experience coding, so i know what i am reading). I find v...

https://x.com/i/status/2070181740094566428

...and was anyone expecting it as such? Anthropic themselves said it would have been pulled regardless around June 22. This ship has long sailed. source

I know NVIDIA and then Samsung/SK HYNIX/MICRON take some heat but in terms of public perception it just seems like everyone piled on AI companies and not the manufactureres that raised prices .... obviously both are responsible also this is just for fun

Is this the Move-37 moment for flooring? I know, this machine is engineered for this job and probably needs close to perfect conditions to work, hence lacking the "creativity" of AlphaGo. But still, don't look where we are today, but 2 more machines down the line.

Maybe the more interesting question is whether true neutrality is even possible for AI systems trained on human data. Should an AI: Present all major viewpoints equally? Follow expert consensus, even if it overlaps with one side more often? Explicitly argue both sides every time? What would you consider a genuinely neutral AI?

I’ve not come across any public reports to this effect, but perhaps others in the group have: Have any AI researchers built an AI system in an air-gapped, secure box, and then told it to take all available steps to break free? We keep seeing layperson-level articles about AI systems supposedly breaking into things like NSA networks. What about AI systems breaking out?