Key insights
- The article argues that while AI LLMs are perceived as a threat to traditional software companies, their current limitations in error detection and reliability, coupled with a dependency on cloud infrastructure, present opportunities for established software firms. These companies can integrate LLMs into their existing robust systems, offering more dependable and potentially offline solutions. This suggests a potential shift where software companies leverage AI rather than being replaced by it, impacting tech valuations.
I see that AI LLM models became a transversal tool in our everyday tasks. Many software companies suffer stock price impact due to this perception that data input in a Claude, ChatGTP or Gemini substitutes software that previously executed tasks seemly but at subscription price. We become mesmerized with the visual impact of these new tech tools but…
There is a big but.
These are LLM, they learn from data they collect and extrapolate to similar situations. They can’t detected errors in the data input. If you feed the model with wrong information, the results will be wrong because the process will go wrong.
A software company produces a tool. A LLM produces a workflow. I can give an example in the medical area.
Ask the interpretation of a chest X ray. The LLM will give you back a well detailed report of many similar X-rays reports based in info it got from previous users. But if you doubt about the result and express skepticism…the model will most of the times try to move the final interpretation towards yours.
For a reason we have erasers and pencils. Redundancy is a natural brain backup system to execute tasks. LLM models are accelerators but they need to be programmed to become efficient. You end up in another piece of software based in GPU, paid subscription instead of a CPU, paid subscription.
This moves back to the early idea: software companies are the ones that can give an immense advantage to the user by incorporating LLM process. They have all the redundancy to verify results, LLM outputs no. Saab or On-premise will be always the most reliable tools. In cases where network is suppressed by any reason, humankind cannot rely in a center that is thousands kilometers far. You need the help of the laptop or computer at your fingertips.
LLM speed up our life by they lack reliability in the outputs (can be improved) and don’t give you off grid independence (you need a datacenter connection).
I think that software companies will move back to their proper value once this LLM fever is gone.
I would like to have your opinions and ideas about the prospects of investing in beaten down software stocks.
PS: not a AI post, created with help of an Ethiopian expresso.