I think way too much is being read into this patent.- The USPTO...

I think way too much is being read into this patent.

- The USPTO doesn't do much to review patents -- they expect that the process to invalidate a wrongly granted patent is for someone to challenge them

- The patent to me looks overly broad and full of things that either had been done before (i.e. prior art), obvious, or amount to "do something from the real world on a computer" none of which are patentable

- The value of a patent is either defensive or exclusive

- Defensive means if someone sues you for infringing their patent, you have the potential of asserting that they in turn are infringing yours and negotiating a mutual licensing agreement

- Exclusive means that no-one else can do what you're doing without licensing your patent. But you need to assert that against someone (i.e. spend a lot of money on lawyers) and run the risk that the result will be your patent being invalidated

For what it's worth I can find no patents at all filed by Dubber itself which is a pretty damning indictment of the last 12 years of alleged R&D that they've been doing. The briefest of searches turns up a lot of patents about meeting summarisation - from companies including IBM and Microsoft - which predate this one.

And beyond the patent, as c1sco pointed out the whole approach to solving natural language problems has changed completely with the rise of LLMs, enormously lowering the barrier to entry to doing this stuff. That means the techniques in this patent - if indeed there is anything novel - are unlikely to be interesting to anyone in 2024 anyway.

In case you're wondering what that means:

- before: a team of engineers and data scientists spend 12+ months hand annotating data and training models to specifically do the task you want (e.g. meeting summarisation)
- now: prompt ChatGPT or another LLM with "here is the transcript of a meeting. produce a summary in 200 words"