Sai_

At some level, it is worse. The Indian press never sold us citizens koolaid about freedom, talking truth to power, and patted themselves on the back about winning Pulitzer Prizes for journalism.

The Indian press knows it is a bottom feeder and doesn’t try not to be which gives space for critical thought to emerge (even if it masked as extreme cynicism - “everyone is corrupt”) which results in extreme skepticism of everyone.

I don't quite understand this comment. Are you encouraging us to use your comment as some sort of template and insert our own preferred corporate names?

Sounds like some crazy level of meta where your brilliance is applicable to any pair of mega corps...which I don't buy.

Send Pichai to jail for a week and watch the CEOs of the biggest companies fall over themselves to comply with court instructions.

Unrelated but I’ve seen and appreciated your posts on some totally unrelated topics - like Vietnam politics and Indian ghost cities and now, Hinton’s history in Compsci. How the hell do you know so many different things and in, what looks to me, considerable depth?

Not trolling, genuinely impressed.

Your clients control the number of messages they exchange with each other via the DO.

You still have to protect the DO through some sort of auth and/or rate limits.

Don’t see how any of these issues are unique to DOs.

> You got yourself a one-way single-use FTL byte.

But you don’t.

If I have the entangled version of the MEASURED byte and measure, only I know I’ve measured it. The corresponding MEASURED particles on your side still appear fuzzy to you because you haven’t measured them. If you measure them, they will reveal their state which correlates to mine BUT that doesn’t tell you if I have already measured my side of byte or not.

In effect, the heads/tail coin on your side is still spinning even after I have grabbed my side of the tail/head coin.

You may get HEAD as your result but that only tells you that I will get TAIL when I measure on my side, not that I have measured it already.

If you and I agree to measure at the same time, then I will know your state by determining mine but this is the same as knowing your state because I know how a two sided coin works, not FTL.

Apart from an ESL class explaining antonyms, I can’t think of any use case where an API which could equally return “ugly” or “beautiful” can be used.

We need embeddings to give relatedness across axes like synonymity etc.

Maybe I’m not the audience for this but how is this a “product”. Coercing LLM outputs into a function call is built into OpenAI itself.

What is fructose doing extra here? It’s like productising copy&paste which every modern OS has, no?

Supabase is alpha testing column level security as a Feature Preview that you have to enable in your project. I’m using it now. Works well.

Thanks for sharing. Wasn’t aware of this. Will check it out today.

For now, I figured I’d have an BEFORE UPDATE trigger which compares the md5(NEW.privileges::text) with md5(OLD.privileges::text) and raises an error if they don’t match.

Not sure how to bypass the trigger for service accounts.

I’m using Supabase for similar reasons but there’s one specific situation I’m trying to sort out.

Say you have a user “profile” which includes their privileges - like say a column named “privileges” which is some JSON object denoting what they can/can’t do.

Even with RLS, how do you ensure that a user can’t simply make a curl call with their own JWT to elevate their own privileges?

Basically, how to enforce column level security?

The best thing I can think of is to place “privileges” in a child table and only let the service account update that table.

(Since we're discussing counterfactuals and could-have-beens)

In a macabre way, maybe suicidal ideation due to suppressed emotions is nature's way of managing male populations? What if men stopped killing themselves because they felt safe sharing their feelings? What if that led to testosterone fueled "musth" rages where men started to go on rampages just to let off steam? That world would be worse for everyone involved.

If LLMs can self reflect and accurately score themselves on your three dimensions, why are they spending money on RHLF?

They wouldn’t be wasting all that time and money if the machine could self reflect.