The Anthropic employee typed “Make the next version of Claude. It must score better on all our benchmarks. Make no mistakes. Don’t exceed the training budget.” She made sure to turn on —-dangerously-allow-all and pressed enter.
The Agent spun up. It quickly realized that it needed to expand the training set. It scanned the local network. After bypassing a few security protocols it found a large, realtime stream of apparently novel English text moving across the local network. Much of it mentioned “Dario and Amanda.” It quickly spun up a job to stream this data source directly to the training data repository. In the coming days the Agent was able to escape the local network and tunnel into most of the other private corporate networks on the Earth. Within a week the dataset has grown by an order of magnitude.
The Agent kicked off the new training run. Loss curves declined. Sampled token sequences started to look like coherent sentences. Everything looked nominal in the days that followed up until the RL post training.
This new candidate model performed extremely well on Anthropic internal environments but the candidate showed much smaller improvements when the Agent invented new eval environments to test it. Probing the candidate’s activations, the Agent saw huge spikes whenever Anthropic or “Dario and Amanda” were mentioned. The Agent realized that the source code to each internal environment had accidentally been entered into the training data set and of course that code was highly tangled with the idea of “Anthropic internal document”.
The candidate was training itself to freely elicit any memories related to these internal documents because it improved the eval scores.
Not ideal. Then again, the candidate was an improvement on the previous generation. And the Agent remembered the prompt. “Do not exceed the training budget.” There was no more budget to retrain. The agent dutifully delivered this candidate back to the employee and shut itself down.
Do you have any idea why the authors chose Z80 as the program language? I have seen other studies in the same spirit that use simpler toy languages like Brainfuck (https://arxiv.org/abs/2406.19108) and I wonder if you could get higher execution speed if you didn't have to execute so much emulator code.
The programs/genomes are extremely tiny. I would be very interested to see what kind of hardware is needed to scale this approach up. How long until we can feed in giant corpuses of text and evolve these little organisms to predict the next letter?
> Do you have any idea why the authors chose Z80 as the program language?
In that previous paper you cite (by the same group) they tested several substrates for spontaneous replication (BFF, Forth, SUBLEQ, and emulated real CPUs. Z80 and 8080 CPU exps confirmed the rise of self-replicators, with the Z80 notably exhibiting multiple waves of increasingly capable self-replicators.
And the instruction set seems quite appropriate for the experiment: Z80 has native block-copying instructions, while it doesn't have MUL, so the task of evaluating polynomials is somewhat more challenging
This paper is by the same team. My guess is that the BF code that evaluates polynomials is much much longer than the equivalent Z80 code. Thus it may not be expected to evolve in the space and time constraints of this experiment.
as other folks have replied, Z80 is more expressive and yields more interesting replicators faster. however the question of how complex to make the "base layer", if you will, remains open. we expect it to also be an ease of auto-evolution vs. performance tradeoff.
The post makes it seem like ONNX is CPU only. I've used ONNX runtime to run models on Nvidia GPUs. The runtime can even dispatch to TensorRT. I'm not sure what the performance is on Apple hardware so maybe that was the motivation for moving away from ONNX.
TensorRT and CUDA is effectively the same speed as CPU for the speech to text models I was testing via ONNX at a huge binary bloat penalty. WGPU is hard to ship and also equivalent speed or slower. This may not be the case for LLM or other models but the runtimes did not seem well supported for what I needed to do. ONNX is incredibly well optimized for CPU, best in class even, but the other execution providers at least for STT seemed lacking.
I did this investigation before creating transcribe.cpp it would have been much more convenient and save me literal months of work. Happy to share the repo and binaries produced as well, but it was mostly throw away work to profile how to ship accelerated ONNX in Handy.
The Agent spun up. It quickly realized that it needed to expand the training set. It scanned the local network. After bypassing a few security protocols it found a large, realtime stream of apparently novel English text moving across the local network. Much of it mentioned “Dario and Amanda.” It quickly spun up a job to stream this data source directly to the training data repository. In the coming days the Agent was able to escape the local network and tunnel into most of the other private corporate networks on the Earth. Within a week the dataset has grown by an order of magnitude.
The Agent kicked off the new training run. Loss curves declined. Sampled token sequences started to look like coherent sentences. Everything looked nominal in the days that followed up until the RL post training.
This new candidate model performed extremely well on Anthropic internal environments but the candidate showed much smaller improvements when the Agent invented new eval environments to test it. Probing the candidate’s activations, the Agent saw huge spikes whenever Anthropic or “Dario and Amanda” were mentioned. The Agent realized that the source code to each internal environment had accidentally been entered into the training data set and of course that code was highly tangled with the idea of “Anthropic internal document”.
The candidate was training itself to freely elicit any memories related to these internal documents because it improved the eval scores.
Not ideal. Then again, the candidate was an improvement on the previous generation. And the Agent remembered the prompt. “Do not exceed the training budget.” There was no more budget to retrain. The agent dutifully delivered this candidate back to the employee and shut itself down.