From AGI to SSI focus for Language Model companies? No GPT 5 it seems

From Abhivardhan, our President

This post on X has so many nuggets that I cannot miss discussing.

Was Gary Marcus right all along? YES, and yet some after-effects that this account has discussed, require some group-think here.

1️⃣ OpenAI's o3 model had severely failed, and rumours were spread that a GPT 5 from scratch could be built. Now, I don't know but the contamination of post-training data output that is juggling inside these large language models and GPTs is creating more problems. I was discussing with Mohammed Soliman some time back how these models had reached bottleneck.

2️⃣ AI alignment has failed as a AI safety/ AI governance concept, and nobody who is a law practitioner, or a policy practitioner should take this branch of AI safety seriously. Since there is no major algorithmic breakthrough beyond Group Relative Policy Optimization (GRPO) in reinforcement learning - this was also obvious.

So what's next?

I think language models can be used for simpler tasks, and they won't lose their relevance. However, the influx of talent being distracted to LLM research and product development, will have to be pivoted to better opportunities.

And it's fine for xAI to focus on SSI instead of artificial general intelligence. That's a good herring.

Btw, interesting that Mike Brundage (ex-head of international affairs at OpenAI) also finds that people are more interested in AI Triage approach in governance-policy (which Sarah Clarke had predicted on LinkedIn), so, I am kind of not surprised.

Interesting times. What do you all think? I think Dr Marcus is vindicated.

Meanwhile, I had defined a term - GaryMarcus'd, which you should check out at https://www.indicpacific.com/glossary-f-j/garymarcus'd