An upside to fierce competition at the frontier is: it should reduce the chances of AI labs withholding powerful models and using them to accelerate capabilities internally. But like you’ve said, how long this dynamic will hold is something worth pondering.
For some reason—e.g. compute scarcity, increasing r&d costs, and possibly government interference—I think the AI “frontier” will eventually become more concentrated/centralized than it is now, thereby weakening competition that might discourage labs from hoarding their best models.
i think our paradigms have to evolve as fast as possible. it shouldn't be open source v. proprietary anymore. goal is to build positive ai & safeguard humanity so we need the best possible recipe of methods to solve it
for example, if we push for open source, it means more dev eyes on how it is developed, bugs chased out, faster dev't. but very easy for a malevolent coding agent to fork & use for nefarious purposes.
I agree. Relatedly, open-weight models by default catch up with the current frontier in roughly 12 months; curious to see how the tradeoff will be governed, since the best models today are adjudged to already be useful for bad actors who want to develop bio weapons.
An upside to fierce competition at the frontier is: it should reduce the chances of AI labs withholding powerful models and using them to accelerate capabilities internally. But like you’ve said, how long this dynamic will hold is something worth pondering.
For some reason—e.g. compute scarcity, increasing r&d costs, and possibly government interference—I think the AI “frontier” will eventually become more concentrated/centralized than it is now, thereby weakening competition that might discourage labs from hoarding their best models.
i think our paradigms have to evolve as fast as possible. it shouldn't be open source v. proprietary anymore. goal is to build positive ai & safeguard humanity so we need the best possible recipe of methods to solve it
for example, if we push for open source, it means more dev eyes on how it is developed, bugs chased out, faster dev't. but very easy for a malevolent coding agent to fork & use for nefarious purposes.
I agree. Relatedly, open-weight models by default catch up with the current frontier in roughly 12 months; curious to see how the tradeoff will be governed, since the best models today are adjudged to already be useful for bad actors who want to develop bio weapons.