genidoi

15-20 sec into any of the demo recordings and I get the sense of 'noone is really there' from the AI responses. I guess it's to be expected from a v1 where the scope is prioritising how the demo conversation flows rather than its actual substance.

To admit having messed up they would need to include the full private key in the post, which they seem to have decided is still more embarrassing than not including it.

> And that skill is?

It is easy to argue otherwise, but to request a response in the form of a web app / embedded widget is at least some kind of AI proficiency that mirrors past experience having used a tool.

Even if you don't agree with the sentiment it's not hard to understand. The pace of AI improvement has strictly accelerated, and strict acceleration is likely going to be the way things go from here. To many, this means mourning a steadier future that is no longer likely to happen.

It's not a benchmark, it is a meme benchmark.

It’s obviously not about the cost of the parts, there just can’t be any tolerance for criminal conduct in government agency employees.

> RSI is obviously

There is no publicly known reference example of RSI, we have no idea how it works or what it does to the trajectory of progress?

No it’s not the same thing because one serves food, and the other facilitated academic fraud.

That approach doesn’t scale though.

Good code reliably transforms real world state in a desirable way.

It's probably quite difficult to manipulate per-user metrics after the data is collected.

At least the gun was smoking and not the foot kind.

Why would they have more spare capacity than Anthropic? Do you have a source for that?

It's frustrating when I catch it doing this because it's wasting tokens and time on what it reports as "one-off" scripts. Might add a requirement to never use shell scripting unless the task is truly a one liner.

I find LLM's too fall into the trap of writing a bash script for a task that clearly needs to be implemented in an Actual Language with Real Data Structures. For example, ask an LLM to bring up a SQL server with some schema + data preload step, and it will write a profoundly long bash script to do that task, every time.

> you can use AI to understand something and map it to your own mental map

This "symbiosis" (for lack of better word) of human with AI seems to be an emergent value proposition of AI. In the process of doing stuff with AI, producing artefacts like code diffs, we are continuously able to decide how strong the mental map is of the current stage of the production process.

I could probably have worded this better but I'm sure it's something others have noticed... this choice we are able to make of how high fidelity our own understanding needs to be of the current working problem, and how that choice never really existed prior to AI.

> However AI cannot meaningfully handle feedback and learn.

Well this is the central bet of AI coding isn't it? We, the humans-in-the-loop, get better at knowing ahead of time which patterns AI will handle better than others, all the while the models actually get better.

Right? The last month alone has felt like a full year of experience compressed into 30 days.

> Starship at $170B is pure option value on technology still in advanced testing.

The argument that Starship is somehow an experimental/unproven technology that might fail to materialise was absurd but plausible sounding before flight 1, there were many new technologies simultaneously being deployed to a single launch system in one go.

But after 3 tower catches of the booster demonstrating centimetres of guided precision of the entire stack, this is becoming a tired argument.

I know the author is not making that case at all here, but it seems like one the core reasons to undervalue SpaceX is that Starship might not work out, and this all sounds exactly like how reusability might not work out for the Falcon 9 from 10 years ago.

Oh it is yeah.

> So I asked him. "What is your developer workflow using Copilot?" I was not prepared for the answer he gave me:

I don’t know why I get annoyed when LLM’s and their output are casually referred to as “he/she”, particularly by non-techies, but I do. There’s something about personifying an LLM that seems incorrect. Perhaps it’s a fear being stoked that increasingly, people might actually be thinking of LLM’s as living beings.

We're speculating here but between time correlation, browser fingerprinting and telemetry, the average user attempting to pull of a clean compartmentalisation of two accounts has no chance, even when they think they do.

Makes you wonder if Google models how much revenue they lose specifically to this fear, because it's very real. I would simply never use GCP or Gemini because the idea of being banned from Google for absolutely any long-tail reason is a far greater cost than any benefit I could derive from those services.

There is realistically no way to evade the account correlation systems that Google (likely) has.

That can be solved by filtering out any posts made after November 2022.

Those are all arguments for why Ethereum is a bad cryptocurrency, not for why Ethereum isn’t a cryptocurrency at all.

Especially given the LLM does not trust the user. An LLM can be jailbroken into lowering it's guardrails, but no amount of rapport building allows you to directly talk about material details of banned topics. Might as well never trust it.

The code example that the author gave was cute:

> spoiler alert, the answer is `df["A"] = pd.Series(a,dtype='object')`

A whole article hinged on rudimentary pandas API knowledge that doesn't even matter anymore with agents.

No vigilant insider is making a series of "single market predictions with high accuracy" on the same account. They would make unlinkable bets on fresh accounts.

You can't be sure that they are an insider or lucky, just from onchain data.