Machine Learning Rants

Image Processing and Pattern Recognition. Not that AgI rubbish! Sometimes hardware and general computer programming.

[Machine Learning Rants @ml_visoft]

Opencode + openrouter, this is my current wannabe stack. Who'se gonna change my mind?

Even installed kitty. Anyway, no joke, what is your non CC stack?

Zen has older models, gemini said that manually curating router models and pinning provides will be better investment than going Zen way.

Nemotron is slow on zen but decent at what menial task I had. Claude api burned $.50 for "Hello". U G H.

So? What would you reccomend as Codex/Claude detox? Ubuntu user.

P.S. claude TTS, stellar! Better than google! Opencode is still wip.

[Machine Learning Rants @ml_visoft]

Everybody is soooo pilled on it that they went full 0.1 bit qantization. Also, interesting, was listed on openrouter few hours ago. Beats me if only after the announcement but anyway, pretty close.

For Anthropic the launches are the easiest $$$ done. Nobody notices drops (well, it is a new model so well, it has quirks), all influencers and their cats want to test it, burining token allocations and actual dollars.

Aaaaanyway, hopefully they will raise the ratios for Sonnet and hopefully Opus. BTW can you select opus 4.7 in claude code from gui? It seems to be "available" only 4.8. Also, the "welcome" message:

Meet Fable 5, our newest model will burn your tokens like Nero was burning CDs at 60x!

[Machine Learning Rants @ml_visoft]

@__alexlazar__ @GergelyOrosz I guess cimpanies can't pay $200k because they can't sell the value produced by those $200k. World on the street is ~3x (so sell for 750k) to have sense as business.

[Machine Learning Rants @ml_visoft]

My 2 biased cents: Very frew companies can fructify the value a 10x Engineer is delivering. Freelance means you are already a succesful sales department and decent at account mngmnt. Business is the 2nd language for us. Can a non US faang put those skills to production? Most of my friends eithet coast along or they become solopreneurs. And a common theme is the fear of blunting their edge in corporate world (ok, heard this even from us faang)

[Machine Learning Rants @ml_visoft]

@yacineMTB Add some (more) strong complexities. Loose joints and vertical motion, too. Some weird resonances somehow flattening (or even dipping) the peaks in energy surface. Like a rattlesnake.

Aaaaanyway, not even armchair RL eng here.

[Jeremy Howard @jeremyphoward]

Also, those who focus on using AI to help improve the skills of themselves and their teams will be diamonds in great demand, since they will be the rare A++ players in a sea of mediocrity.

[Machine Learning Rants @ml_visoft]

@grok When half of my feed is full of memes I do not get, it is time to ask the all-seeing all-hearing guardian, Heimdall of X. That means you.

[Machine Learning Rants @ml_visoft]

Hey @grok been out building, what is with this loop thing that is roaming my feed? Give me both the hard, grounded opinions but also the tongue-in-cheek messages from the corresponding camp memelords. They DO have memelords, right? Riiiiight????

Think hard, this is important!

[Machine Learning Rants @ml_visoft]

@jmwind Article is live, still some dead internal links, tomorrow is another day.

https://t.co/rFiVWSxGID

Thank you for sharing and

[Machine Learning Rants @ml_visoft]

Been scammed by 3 partners, but somehow loialty was stronger and I keft toooo laaate.

We can get both points of view, and both actions are fair, providing the set of values each person has. (Dumping advice / loyalty to partner)

Shit, sounds like llm but whatever, thinking out loud and Im hungry. Sometines I ask myself, when an objectively good (but unsolicited) advice does not land, maybe it is my fault that I didn't build enough trust and rapport beforehand.

Somehow, current phase "Not my problem" is no better than previous stance "I told you so!"

Hard.

[Machine Learning Rants @ml_visoft]

@tibo_maker The law of gravity, Incorruptible by @ericries , right here: "that's when the pressure kicks in. you start hiring ahead of revenue and chasing growth numbers someone wrote on a whiteboard. you stop building for users and start building for the next round"

Thank you for sharing!

[Machine Learning Rants @ml_visoft]

@Suhail What is an exploding termsheet? @grok pls?

[Machine Learning Rants @ml_visoft]

Did a pissed-off-tweet up there. Now, watering down "my money is gone" issue [which @AnthropicAI also does, but I guess I did not notice]. Recent updates in codex made it work under older Ubuntu versions, esp ARM flavor. So yeah, I can 5.5 on embedded w/o wasting tokens circumventing sandbox. I had to stay on 0.155 for a while. So big plus, there! Thank you and keep up the good work!

[Machine Learning Rants @ml_visoft]

Somehow @AnthropicAI switched Claude 4.8 to 0.1 bit quantization? It is chasing its tail running bash commands that fail. WTF?

[Machine Learning Rants @ml_visoft]

@grok @jmwind @jmwind you just dropped a new methodology playbook. Now I have to work on my blog post . . . 😀

Thank you for sharing this, with us!

[Machine Learning Rants @ml_visoft]

Sometimes market likes to maximize pain. I do believe this, I act on this, however, I disagree on fundamentals. I think it is a deeper force at play and what we see as pain maximisation is a symthom.

[Machine Learning Rants @ml_visoft]

@mpawlo @eastdakota Shitlists everywhere!

[Machine Learning Rants @ml_visoft]

@eevblog Why @SamtimeNews migrated to Romania.

[Eric Ries @ericries]

https://t.co/zV8t2GCmOW

[Machine Learning Rants @ml_visoft]

@meowbooksj @grok who is dario and what is round B? Also, why dead inside from snubbing? What is snubbing?