It's simple for parents - be the human tutor they need. If you're capable of loving them and being responsible for raising them, that task fall to you. Underprivileged by socioeconomic realities? In the US there is Head Start for this problem, and yes, it does work and has been longitudinally proven to do so by independent studies and peer reviewed publication. In a nation without these services, but able to be here on HN reading about the situation? Who is helping your child right now?
The concept of using untested, unregulated, demonstrably harmful models - see news about 'suicide ideation and school shooting facilitation' by public facing enterprise models - is a dangerous condition that no amount of hand-wavy, but "My co-founder, who is a clinical child psychologist" statements mitigate.
First, that isn't a statement by the actual person purporting to be a professional, and second this statement plays into the misconception in American consumers' minds which often conflates psychiatrist with psychologist . The first is a medical degree with the responsibility and risk of losing a medical accreditation for engaging in this kind of unethical, untested approach on children at large. The second has no supervisory board capable of revoking credentials, and may even mean nothing more than an undergraduate degree in basic psychology. Thus, I am highly skeptical about any such business' ability to carry liability insurance from a reputable firm.
Here on HN you can readily find articles about firms offering fraudulent compliance readiness; so I'm not doubtful of being able to procure the verisimilitude of coverage, just the veracity and legality of the coverage available to any organization taking this cavalier approach which relies on these models.
Now, if the model and infrastructure Elo is leaning on is an in-house, on-premises, refined private MoE (Mixture of Experts) framework with a curated harness and training data which is demonstrably free of harmful materials and has been/continues to be fine tuned to provide provably safe content and interactions: that might be testable for compliance and liability coverage.
The cost of such an infrastructure would make it a premium product, with a potential customer base that can and does already have 'affordable' human tutoring. As this human one-on-one to on-on-a-few tutoring is the gold standard benchmark for 'best-case' that this product cannot exceed, there is no viable use-case for it's existence or use, except when the parents might find an AI subscription to be less of cognitive load than managing actual people. Even rich people can be self serving and delusional about the well-being of their own children.
All I can see from here is; attention grabbing, AI hype delusional, high risk, for-profit exploitation of overworked middle class parents, using uncontrollable public facing models which hide behind a 'we take no responsibility for the results' disclaimer organizations that can and will spin the dials on performance and capacity quotas and 'upgrade' to a degraded, de-safety limited model with zero notice, mid workload. Which once the enticingly low costs of providing this 'risk factors as a service' are covered, has the appearance of profitability for the founders in the short term, and when the 'unlimited plan' is actually very limited will be an external 'force majeure' for either subscription price increases or rug pull from under the by-then addicted young minds.
The need for solutions to the gap in educational opportunity between children of the top 1% and the rest of the world is real. This approach is not a solution; but is an entirely different class of risk posing as a solution.
I do enjoy a good stream of consciousness rant, so thanks for that.
A simple rejoinder - The OP is one solution to many households where they can just be "the human tutor". If you can't see that, you simply don't know the average household.
Here's are a few links to several resources about the state of safety at the frontier LLM models' companies of origin. The first uses shorter sentences and has further supporting links.
Hopefully those with provide you with an easier to comprehend, "synthetic" (sic) measurement of the risks that are unavoidable in the 'product' under consideration.
TLDR; the safety at every commercial LLM scores at or below a C+. I don't believe anyone serious can honestly believe that the founders of this 'product' are capable of reinforcing security that does not exist in any of the models they may choose to utilize.
Also, the "Shame on you" statement was directed exclusively at the founders of this product and people who would intentionally expose children in their stewardship to these dangers.
If that is not you, then my honest apologies if it seemed directed at you. That was not my intent.
If that includes you, and you couldn't grok the concepts in my rant; we have nothing more to talk about.
The concept of using untested, unregulated, demonstrably harmful models - see news about 'suicide ideation and school shooting facilitation' by public facing enterprise models - is a dangerous condition that no amount of hand-wavy, but "My co-founder, who is a clinical child psychologist" statements mitigate.
First, that isn't a statement by the actual person purporting to be a professional, and second this statement plays into the misconception in American consumers' minds which often conflates psychiatrist with psychologist . The first is a medical degree with the responsibility and risk of losing a medical accreditation for engaging in this kind of unethical, untested approach on children at large. The second has no supervisory board capable of revoking credentials, and may even mean nothing more than an undergraduate degree in basic psychology. Thus, I am highly skeptical about any such business' ability to carry liability insurance from a reputable firm.
Here on HN you can readily find articles about firms offering fraudulent compliance readiness; so I'm not doubtful of being able to procure the verisimilitude of coverage, just the veracity and legality of the coverage available to any organization taking this cavalier approach which relies on these models.
Now, if the model and infrastructure Elo is leaning on is an in-house, on-premises, refined private MoE (Mixture of Experts) framework with a curated harness and training data which is demonstrably free of harmful materials and has been/continues to be fine tuned to provide provably safe content and interactions: that might be testable for compliance and liability coverage.
The cost of such an infrastructure would make it a premium product, with a potential customer base that can and does already have 'affordable' human tutoring. As this human one-on-one to on-on-a-few tutoring is the gold standard benchmark for 'best-case' that this product cannot exceed, there is no viable use-case for it's existence or use, except when the parents might find an AI subscription to be less of cognitive load than managing actual people. Even rich people can be self serving and delusional about the well-being of their own children.
All I can see from here is; attention grabbing, AI hype delusional, high risk, for-profit exploitation of overworked middle class parents, using uncontrollable public facing models which hide behind a 'we take no responsibility for the results' disclaimer organizations that can and will spin the dials on performance and capacity quotas and 'upgrade' to a degraded, de-safety limited model with zero notice, mid workload. Which once the enticingly low costs of providing this 'risk factors as a service' are covered, has the appearance of profitability for the founders in the short term, and when the 'unlimited plan' is actually very limited will be an external 'force majeure' for either subscription price increases or rug pull from under the by-then addicted young minds.
The need for solutions to the gap in educational opportunity between children of the top 1% and the rest of the world is real. This approach is not a solution; but is an entirely different class of risk posing as a solution.
Shame on you both.