Bleeding Heart Liberal

Data Scientist with 50+ years of R&D. 45 years of happy marriage. I do not respond to DM's. Everybody here seems to hate everybody else; what's up with that?

[Bleeding Heart Liberal @BleedingHartL1b]

@MahikRani50377 $1.60 was the non-farm federal minimum wage when I started working.

[Bleeding Heart Liberal @BleedingHartL1b]

@Dviddles @WallStreetMav Reagan was a disaster for America.

He gave away our manufacturing base for a few decades of cheap labor.

[Bleeding Heart Liberal @BleedingHartL1b]

@WallStreetMav STEM majors still have value. No way I would have learned the math I got from my Math Bachelor's by self study.

[Bleeding Heart Liberal @BleedingHartL1b]

@Capitana_espana Dismal Oubliette Seating.

What it really needs is a box and a lid for each passenger.

Passengers can wear Depends Adult Diapers rather than needing passage to the toilets.

[Bleeding Heart Liberal @BleedingHartL1b]

@RpsAgainstTrump So a South African fascist is lecturing us now on who is an American?

Why don't you actually get FSD working instead of posturing or something.

[Bleeding Heart Liberal @BleedingHartL1b]

@EdGeeDetroiter @johnrich You are correct, Ed.

I was about to write that, but now I don't need to.

Thanks for solid factual inputs.

[Bleeding Heart Liberal @BleedingHartL1b]

@Indy_triguy @NVIDIAAI Actually, there is. They have weights that can execute code that phones home. A sysadmin sandboxed Qwen and monitored its network activity. Sure enough, it phoned home to the CCP.

Plus, you can bet they are doing it if they say it in the terms of service.

[Bleeding Heart Liberal @BleedingHartL1b]

@TashaViolet23 @bennyjohnson Oh, so you oppose pedos now?

When did that start?

[Bleeding Heart Liberal @BleedingHartL1b]

@Indy_triguy @NVIDIAAI One thing to be aware of is the Qwen models Terms of Service flat states that everything us do is logged and sent to the CCP for analysis and distribution to Chinese industry.

Gemma, Llama, and Mistral, local models are much safer to use if you have any IP to protect.

[Bleeding Heart Liberal @BleedingHartL1b]

@Indy_triguy @NVIDIAAI I am just using Ollama, but recent upgrades that have dropped noticeably speeded up inferencing. Do you have the latest version?

[Bleeding Heart Liberal @BleedingHartL1b]

@Indy_triguy @NVIDIAAI BTW, I get 69.74 tokens/sec from 26b on my gaming rig.

[Bleeding Heart Liberal @BleedingHartL1b]

@Indy_triguy @NVIDIAAI Just ran gemma4:26b MoE on my Raspberry Pi 500 Plus and got a 2.7 tokens/sec eval rate. The larger dense gemma4:31b and it pages into oblivion. So 26b is the biggest model it can run.

[Bleeding Heart Liberal @BleedingHartL1b]

@Indy_triguy @NVIDIAAI Just reran devstral-2:123b at 1.02 tokens/sec. It requires patience, but it runs. It is still faster than writing the code line-by-line. The trade-off is you make fewer runs with the bigger model.

I started programming with punched cards, which REALLY required patience.

[Bleeding Heart Liberal @BleedingHartL1b]

@Indy_triguy @NVIDIAAI I am trying out nemotron-3-super:120b right now. The 30b class models are for edge devices and pre-AI laptops.

[Bleeding Heart Liberal @BleedingHartL1b]

@lukesmithapps @shashj Yet the AI forward companies are the ones doing more hiring these days.

[Bleeding Heart Liberal @BleedingHartL1b]

@AllTheTokens @minilek @AnthropicAI You sound confused.

Anthropic people are the closest thing we have to good guys in this field.

[Bleeding Heart Liberal @BleedingHartL1b]

@Riya_XOO1 (36 - sqrt(36))/sqrt(36)

(36-6)/6

30/6

5

[Bleeding Heart Liberal @BleedingHartL1b]

@smsamford01 @nicksortor Only the criminal ones, Susan.

Only with due process, too.