This was really thought provoking from the FT’s AI Shift newsletter. I’m persuaded by Nick Srnicek’s argument that the push for AGI can be understood economically as a search for a product that won’t need to be fine-tuned. But if fine-tuned open weights models can outperform frontier models then this overturns the economics of the AI labs:
But a
[new case last month [email.newsletters.ft.com]]took things a step further, when Ray Dalio’s investment firm Bridgewater Associates partnered with AI platform company Thinking Machines Lab (founded by former OpenAI CEO Mira Murati) to fine-tune a model based specifically on how its own investment managers do their work. As with the legal example, the results significantly outperformed frontier models, this time at a 14th of the cost. Crucially, however, use of the firm’s own proprietary records and its highly specialist staff’s knowhow may make these gains more durable.Bridgewater had its own experts write bespoke prompts that framed questions in a way that guided the models to the correct answers. This produced solid gains, but they still topped out below 80 per cent accuracy.
A prompt only imparts the expertise a professional is able to put into words — what is much better is to learn from their actions. For the fine-tuning step they put together a set of tasks drawn from their own investors’ daily workflows, and crucially also had staff ensure that the ideal responses which would guide the model’s training did not just represent ‘correct answers’ but ‘exactly how our investment professionals would approach this’.
At the end of the process they had a bespoke model whose behaviour had been tuned towards Bridgewater’s own assessment of excellence, taking it up to 85 per cent accuracy — an almost 30 per cent reduction in errors compared to the frontier models — at a tiny fraction of the cost.f
There’s an obvious sociological question here: what happens if our investment professionals need to approach this in a different way? Human professionals can learn and adapt. They can do so in ways that change how they prompt if they’re using a frontier model. In contrast a model fine-tuned in this way will be locked into a particular set of dispositions that might be appropriate for a context at T1 but will cease to be appropriate at a later T2. Indeed introducing the models is liable to change the context, particularly if you use this as an excuse to lay off your staff! There’s an organisational problem here which is blindingly obvious to anyone who has thought in any depth about the fact that contexts change.