I’ve had this happen to me several time over the past weeks and it’s gone from quaint to humorous to farcical to outright “is-the-world-gaslighting-me” insane.
Just today I was reading Stanley Druckenmiller’s op ed in WSJ. This dude is like 80 and has made billions of dollars, and he got Claude to write his op ed???
Unbelievable. And the tells are so obvious, yet people still love the Claude-like quips and odd grammatical choices that read like halfway asshole halfway mid-sentence confusion.
That op ed was absurd. I respect Druckenmiller a lot and am always impressed with his lucidity in interviews. The Claude “ick” was all over his writing.
Note that ElevenLabs claims the right to train on any data you upload to the service, such as editing a song you made etc.
> (d) License to Your Content and NILV. You hereby grant to ElevenLabs a license to use, reproduce, modify, adapt, publish, translate, create derivative works from, distribute, make available, communicate to the public, publicly perform, display, and use your Uploaded Music and Input (including your name, image, likeness, voice rights and other indicia of your persona that may be embodied in your Uploaded Music or Input (collectively, “NILV”)) and your Output, in each case, to provide the Services (including the trust and safety features therein), to promote and improve the Services, and to develop and improve our other services and products. Such license shall be…
They also seem to use a different encoding scheme which has more bandwidth and less artefacts up front. If you download a WAV from Suno, you still essentially have MP3 quality, just serialized to a WAV file. The elevenlabs WAV has full audio bandwidth instead and I checked frequency sections with a parametric EQ and noticed that e.g. the cymbals go high up in spectrum (which is not obvious in the given sample song where middle and bass frequencies dominate); with a bit of mastering the result is much better, but I posted the original here. I can also upload the WAV file if needed.
Formalizing proofs in Lean has gotten dramatically easier since the formalizations available in 2005. And Lean’s mathlib has done most of the underlying work so that you have its axioms and necessary lemmas baked in. You can think in terms of standard abstractions that look very much like the exact notation in the undergrad textbook.
That said, I am not in any way trying to discount how incredible of an achievement it is to formalize a millennium prize winning algorithm in Lean. I mean just look at the code that OpenAI published. It’s like an encyclopedia of different fluid dynamics concepts.
Are you one of the developers ? If so, you need to add a specific page explaining in detail the guarantees that this gives.
For example: point 08 says "Assign different tables or workloads to different shard groups" and point 02 says "Split hot shards as workloads grow". How do those interact ? Can a single table be split across multiple shards ? If so don't you need 2pc to enforce primary key constraints ?
If the agent is able to “decide” when a loop should occur vs when an output token is produced, that effectively moves the CoT inside the architecture. While that’s not what is happening here, it’s clearly a plausible way we could see CoT disappear.
> that effectively moves the CoT inside the architecture
This may be a bit of a nitpick, but... does it? I agree that giving the model decisions on looping certainly makes interpretability harder, because it adds more transient internal states to deal with and changes the number of them depending on prior states. But is it really pulling CoT inside the forward pass, if the sequence length it's operating on isn't growing? In some sense the whole technique and tradeoff of CoT is "add more tokens to the sequence, use them to reason with", with one of the benefits being, you force the model to output tokens, so you can (hopefully) understand it. And the big point TFA is making is, nobody is doing recurrence over sequence length as far as we know.
Just "adding more layers" doesn't explain the step change. We have moved past the point you can just stack more layers and get huge gains from it. Some have called it latent space reasoning.
It is worth noting that None performed better than Low and nearly the same as Medium on ARC3. And with adapter it still scored >96% with no CoT. So I think it is possible but it also cost them more on None.
This is with the caveat that OpenAI uses their own harness for this:
> On ARC-AGI-3, GPT-6 Astra was run with our responses API harness , which changes two settings to better match real-world performance. The changes do not specifically target ARC-AGI-3.
Their neutral harness is not very good though, if I read it right, it doesn't preserve the reasoning state between turns. No real harness discards reasoning state like that.
Yeah that must be it. OpenAI doesn't want to disclose internal reasoning, that's why thats typically encrypted_content in OpenAI codex session ledgers etc.; leveraging responses API preserves reasoning server side all the way till a final answer is made; so that's very impressive and to me the score that matters.
This should be normalised and expected - the responses API harness allows it to use the custom compaction that is not allowed otherwise. It is entirely fair to allow OpenAI to use their own compaction algorithm..
> GPT-6 Astra represents a step-function change in model capability for interactive reasoning problems. It scores 66% on ARC-AGI-3 using our standard harness, and nearly 100% with a continuous conversation harness and custom compaction, at a cost of roughly $360 per game.
> Going forward, we will report both Standard harness and Provider Adapter harness results on the ARC-AGI leaderboard, with each evaluation condition clearly labeled. Our open-source testing repository and testing policy document both approaches.
This is what the Author of the benchmark has to stay. Quality of the comments keep going down smh
"Going forward we will capitulate and still try to keep the integrity of our benchmark in tact, but from now on every benchmark will be compromised with providers being able tweak things sufficiently to game at least a 30% bump in results."
This concept of a free lunch was never true. In a competitive dynamic, speed and performance were always worth optimizing, comparing, and improving.
One of the primary reasons for this is that computers operate in a vast range of orders of magnitude. There’s several orders of magnitude between cache local cpu operation and dram, then several to disk, then several to network, then several to globally durable guarantees. When your code has literally thirteen orders of magnitude to optimize under, there’s never a free lunch. You always need to understand your stuff.
Just today I was reading Stanley Druckenmiller’s op ed in WSJ. This dude is like 80 and has made billions of dollars, and he got Claude to write his op ed???
Unbelievable. And the tells are so obvious, yet people still love the Claude-like quips and odd grammatical choices that read like halfway asshole halfway mid-sentence confusion.
reply