derekmoeller

Founder, CognitionWorks: generative AI to upskill manufacturing

[derekmoeller @derekmoeller]

@mattmireles @Apis_Bicornis @Crispy75 @Monumental_Labs @mspringut You're quoting your own AI-written article back. Stone and brick veneers are common, and these objections are imaginary obstacles. Steel angle irons are used to carry the weight and it works fine. Much better than replacing all the load bearing elements with stone.

[derekmoeller @derekmoeller]

@mattmireles @Apis_Bicornis @Crispy75 @Monumental_Labs @mspringut Why would we use stone for structural parts? We need this tech for the aesthetics. That’s where the market demand is, and the cost of aesthetics drops if you don’t mix it up with the engineered structural part of the building.

[derekmoeller @derekmoeller]

@jmbollenbacher I'm guessing their "introductory price" of $2/10 is giving them optionality to extend that price if they continue to feel the heat from OAI and GLM et al.

This is the first time I recall them offering introductory pricing.

[derekmoeller @derekmoeller]

@SecretCFO Marginal cost thinking is still rare in manufacturing.

[derekmoeller @derekmoeller]

@kristoph @cryptopunk7213 Enterprises are not going to save much or any money by self-hosting a huge model like GLM 5.2, with capital costs factored, vs inference providers.

And GLM 5.2 is token verbose.

But I think you still get to savings, and it'll force labs to stay competitive.

[derekmoeller @derekmoeller]

@arvidkahl For medium or higher value sales, especially recurring, in-person has always been a winning strategy because it's perceived as a cost investment in the relationship and the CLV maths.

It's just more true now because digital tactics are getting more and more undifferentiated.

[derekmoeller @derekmoeller]

@joannejang It's an easier concept to onboard for people, but separate agents with distinct identities that learn and develop skills/memories independently seems like a better long term organizational concept. LucyBot in accounting, BobBot in legal.

[derekmoeller @derekmoeller]

@tekbog I suppose they are betting that (1) having claude code behind it will be a differentiator and (2) entrenched memory will create switching costs.

Seems tough not owning the UI layer though.

[derekmoeller @derekmoeller]

@JulieKallini @jxmnop @jwthickstun Read 19th century English and Russian novels. Better than modern light fiction and breaks you out of the LLM ruts after staring at their outputs all day.

[derekmoeller @derekmoeller]

@dwlz I’m a big fan of AI but this part of it is awful. People just sending unfiltered LLM dumps, often without reading it, and expecting you to do something with it. Feels narcissistic.

[derekmoeller @derekmoeller]

@jmbollenbacher GUIs are not composable. This is like all those companies that tried to kill programming with point-and-click visual builders. It never worked. There's a reason it never worked, but text haters never seem to reflect on it.

Use of GUIs is a human weakness that AIs don't share.

[derekmoeller @derekmoeller]

@NatPurser @tenobrus Aligning vastly different constituencies' interests to achieve a common goal is the defining work of a CEO: investors, employees (senior and junior, sales versus production), customers and regulators all have vastly different interests, and very different philosophies.

[derekmoeller @derekmoeller]

@mitchellh I don't think it matters much if they are local or not, versus being open source. Open source models open up commodity marginal cost inference providers, which cuts token costs by 80%.

Running bigger models locally is expensive and won't be economically useful for a while.

[derekmoeller @derekmoeller]

@MaciekCodes @feliclac @samlambert duckdb/clickhouse

[derekmoeller @derekmoeller]

@iwasrobbed this obviously doesn't work over 10 million PDFs and terabyte-scale data warehouses

good for personal projects though

[derekmoeller @derekmoeller]

@NathanWilbanks_ You’re saying that LLMs will be able to rebuild any proprietary software in the world from binary by the end of this year?

[derekmoeller @derekmoeller]

@stevesi The weird part is that there is not a lot of demand for big data centers inside city limits.

It's performative populism, designed to antagonize the companies that made this area home, without any particular policy objective.

[derekmoeller @derekmoeller]

If you're working on any type of ML -- even non-LLM neural net research -- can you trust Anthropic? It seems increasingly unwise to do so.

There is no way to know if they are intentionally corrupting your work.

Unfortunate to see dishonesty be an official company policy.

[Nathan Lambert @natolambert]

Labs starting to pull up the ladders on the ability to diffuse AI was inevitable. Doing it without telling the user is misaligned.

[derekmoeller @derekmoeller]

@chris_kratovil It’s just labor costs. The marginal cost of being open at 2am went up a lot, and marginal revenue didn’t follow.