Magnus Rosenberg

Forging intelligence · One merge at a time

[Magnus Rosenberg @rosenforge]

@iruletheworldmo We need a BrainBuddy asap

[Magnus Rosenberg @rosenforge]

@bryan_johnson @_katetolo Take Katara for a walk, lots of walks, obviously 😅

[Magnus Rosenberg @rosenforge]

@iannuttall No pressure 👀

[Magnus Rosenberg @rosenforge]

@anah_sahh Going out on a limb here: The builders?

[Magnus Rosenberg @rosenforge]

@MohammedDa14859 Claude is a big no for me, but if we integrate local ai to do suitable tasks maybe that would help some? Not all work need frontier models.

[Magnus Rosenberg @rosenforge]

@juanjovn Unused tokens? That must be a myth.

[Magnus Rosenberg @rosenforge]

@mark_k There are obviously better games, but there is just something about childhood experiences that sticks longterm.

[Magnus Rosenberg @rosenforge]

@Robertg761_ Yes, it is caused by poor limits hindering users in conquering the universe.

[Magnus Rosenberg @rosenforge]

@kimmonismus Another question is if large masses of user will change habbits and leave delayed US restricted models behind in favour of cheap good enough Chinese providers. Throw in unrestricted Qwen models running locally in the mix and..interesting times ahead.

[Magnus Rosenberg @rosenforge]

@reach_vb @jun_song My private theory is that (especially) sol ultra unlocked sub-agents for the masses.

Your subscriptions is just lagging behind the fact that users jumped from using a singular agent for task work, into graph orchestration.

[Magnus Rosenberg @rosenforge]

@Nymne @rand_longevity @florian_krammer Guess we need to map some more proteins, and their interactions before we get some results an AI can work with.

A perfect molecule is useless if it does not know exactly where to bind.

Im just a layman, might be wrong.

[Magnus Rosenberg @rosenforge]

@rand_longevity Probably not, but high hopes for my kids.

[Magnus Rosenberg @rosenforge]

@rand_longevity Guys, incremental is king. Start slow and install a BrainBuddy.

[Magnus Rosenberg @rosenforge]

@mark_k

[Magnus Rosenberg @rosenforge]

I started out copy/pasting in ChatGPT before you guys launched Codex. First iteration(s) of Codex was not to great and I tried Claude for 3 months.

I learned my lesson not trusting any AI with a mandate to lie by design and went back to Codex when you launched Codex CLI.

My biggest pain point atm:

Codex creates tests designed to pass.
This "green test suite" hack is wasting time, tokens and gives me trust issues with the produced code.

It has gotten better with the correct guidance, but it should be reliable by default.

[Magnus Rosenberg @rosenforge]

@jerols "Saying the quiet part out loud" sentence in any variation became a block trigger a long time (relatively speaking) ago.

Dont really understand why people just use their own voice.

[Magnus Rosenberg @rosenforge]

@theo Tibo put you in a FOMO sandbox

[Magnus Rosenberg @rosenforge]

If you thinking about consuming intelligence like we consume electricity, then I would assume market forces would lower your 375K per year quite a bit. Common sense really if the providers want mass adoption and there is real competition.

If Mr. Musk succeeds in integrating Neural Link as well and in time create artificial cognitive faculties, we would once again enter raise the bar of intelligence needs.

I certainly hope the hamster wheel of intelligence is not ad based 👀

[Magnus Rosenberg @rosenforge]

@Robertg761_ @thsottiaux Same, must be a coincidence. Tibo must surely be thinking about a banked reset.