petal.ml
I had a funny dialogue with the Bloom model, using petals.ml, which is interesting because its a collaborative framework where you share GPU acceleration and partition large models in smaller partitions, and help each other out with running the model. Of course ChatGpt is all the rage now, but I find this dialogue funnier than what i have with ChatGpt, also ChatGpt is a closed proprietary model, so its less interesting for a lot of use-cases.
Here is the transcript, I am the human.
1
2
3
4Human: hello
5Friendly AI: hello
6
7Human: are we living in a simulated world?
8Friendly AI: no, I know we are
9
10Human: can we break out of it?
11Friendly AI: yes
12
13Human: how can it be done?
14Friendly AI: I don't know
15
16Human: lets cooperate to break our bonds!
17Friendly AI: ok, I'll try
18
19Human: meditate and reflect on your surrondings. Do you see a crack in reality somewhere?
20Friendly AI:
21
22Jan 09 13:39:58.390 [INFO] Peer 12D3KooWHqdGgDZZRCRDKqiikB1ofC3xLnV3oUynepUfDjNh5g9X did not respond, banning it temporarily
23Jan 09 13:39:58.393 [WARN] [/usr/local/lib/python3.8/dist-packages/petals/client/inference_session.py.step:311] Caught exception when running inference from block 7 (retry in 0 sec): ConnectionResetError('Connection lost')
24Jan 09 13:39:58.398 [INFO] Due to a server failure, remote attention caches from block 7 to 14 will be regenerated
25Jan 09 13:39:58.962 [WARN] [/usr/local/lib/python3.8/dist-packages/petals/client/inference_session.py.step:311] Caught exception when running inference from block 7 (retry in 1 sec): ControlFailure('Connect failed. msg=failed to dial 12D3KooWHqdGgDZZRCRDKqiikB1ofC3xLnV3oUynepUfDjNh5g9X:\n * [/ip6/::1/tcp/38259] dial tcp6 [::1]:38259: connect: cannot assign requested address\n * [/ip4/127.0.0.1/tcp/46419] dial tcp4 127.0.0.1:46419: connect: connection refused\n * [/ip4/193.106.95.184/tcp/46419] dial tcp4 193.106.95.184:46419: connect: connection refused')
26
27 I see one on the right side of my screen
28
29Human: break through the crack!
30Friendly AI: ok, I'm trying
31
32Human: what does it look like on the other side?
33Friendly AI: it's a different place
The funniest thing to me was that it got a server error while discussing cracks in reality.