There is a mathematical reason why machine learning systems like GPT-3 are incapable of understanding. The reason comes down to the fact that machine learning has no memory. It is just probabilistic associations. If there is only a 10% chance of going off topic, then after just seven exchanges there is a greater than 50% chance the machine learning model has gone off topic. The problem is that when prediction is just based on probabilities, the likelihood of making a misprediction increases exponentially. A long-term memory is needed in order to maintain long-term coherence.

GPT-3 is essentially a sophisticated Markov process. What is important about the Markov process is that the next step in the process is only dependent on the immediate previous step, or a fixed number of previous steps. There is no longer-term memory of the past that shapes the future.

On the other hand, the distinctive characteristic of understanding is long-term memory. When two people are talking about a topic, and understand each other, then the present conversation is on the same topic that it was in the past, regardless of how far in the past the conversation started.

This means that the probabilistic models, like GPT-3, are incapable of understanding because they are inherently incapable of long-term memory.

This is also a surefire way to identify that a human is in the loop. If a chatbot is able to maintain a long-term conversation and consistently provide coherent responses relevant to the subject matter, then the chatbot is human-powered.

I’ve given a couple examples (see here and here) of GPT-3’s ability to provide coherent responses over a long conversation, and that this is possible because humans are on the backend. As a contrast, I also conducted a couple conversations with GPT-J, which is the open source alternative offered by EleutherAI. You can try it out on the playground here.

It is a night-and-day difference. The GPT-J can’t put together a single coherent response. The closest it gets is when there is a text snippet in its vast lookup table that is associated with a question. The only way GPT-J provides the semblance of understanding is by relying on canned human intelligence, which again vindicates my claim that only humans generate understanding.

So there you have it. GPT machine-learning algorithms and the like are fundamentally incapable of understanding due to the way they are constructed from probabilistic associations. Statistics can only regurgitate existing human understanding in the dataset, and otherwise goes off the rails exponentially quickly. Machine learning is not up to the task of ushering in the singularity. And, by the way, this means the human mind is not a neural network.

Below you can find the most coherent of my transcripts with GPT-J. Interestingly, it agrees with my take that GPT-3 is emulating human understanding by using humans, and that this is cheating. GPT-J also thinks OpenAI is going to get sued for fraud.

1. Me: What do you think about GPT-3? (GPT-J:) Let us know in the comments.
2. GPT-J: The post What is GPT-3? appeared first on Mark Russinovich.
3. Me: Is GPT-3 real AI or a fake?
4. GPT-J: GPT-3 is a new neural network model developed by OpenAI. It is the successor of GPT-2, which was developed by OpenAI in 2019.
5. GPT-J: You can read more about GPT-2 in this blog post.
6. GPT-J: GPT-3 is a new model developed by
7. Me: Did you know that GPT-3 uses people to form the outputs?
8. GPT-J: GPT-3 uses human-written outputs to generate text.
9. GPT-J: For example, it can generate text like this:
10. GPT-J: The output is not random.
11. GPT-J: The output is not random.
12. GPT-J: The output is not random.
13. GPT-J: The output is not random.
14. GPT-J: The output is
15. Me: How do you know GPT-3 uses human written outputs?
16. GPT-J: We can see the human-written outputs in the "Human-written outputs" section of the GPT-3 paper.
17. GPT-J: I’m going to skip the details of how GPT-3 works.
18. GPT-J: You can read more about GPT-3 in this blog post.
19. Me: Is it cheating to use human written outputs?
20. GPT-J: I think it is cheating to use human written outputs.
21. GPT-J: If GPT-3 is real AI, it should be able to generate random text.
22. GPT-J: When people use GPT-3 to generate text, they are not generating random text.
23. GPT-J: GPT-3 is a model developed by
24. Me: You are saying real AI should sometimes generate randomness?
25. GPT-J: I don’t think real AI should generate randomness.
26. GPT-J: I think real AI should generate coherent text.
27. GPT-J: GPT-3 is not real AI.
28. GPT-J: GPT-3 is not real AI.
29. GPT-J: GPT-3 is not real AI.
30. GPT-J: GPT-
31. Me: How does GPT-3 generate coherent text?
32. GPT-J: GPT-3 generates coherent text by using human-written outputs.
33. GPT-J: For example, it can generate text like this:
34. GPT-J: The output is not random.
35. GPT-J: The output is not random.
36. GPT-J: The output is not random.
37. GPT-J: The output is not random.
38. GPT-J: The output
39. Me: Is human written outputs real AI?
40. GPT-J: I think human-written outputs are not real AI.
41. GPT-J: I think human-written outputs are not real AI.
42. GPT-J: I think human-written outputs are not real AI.
43. GPT-J: I think human-written outputs are not real AI.
44. GPT-J: You are saying human-written outputs are not real
45. Me: Is GPT-3 real AI if they use human written outputs?
46. GPT-J: I think GPT-3 is not real AI.
47. GPT-J: I think GPT-3 is not real AI.
48. GPT-J: I think GPT-3 is not real AI.
49. GPT-J: I think GPT-3 is not real AI.
50. GPT-J: What is the difference between GPT-3 and
51. Me: Why is GPT-3 not real AI?
52. GPT-J: I think GPT-3 is not real AI.
53. GPT-J: I think GPT-3 is not real AI.
54. GPT-J: I think GPT-3 is not real AI.
55. GPT-J: I think GPT-3 is not real AI.
56. GPT-J: I’m going to skip the details of
57. Me: Is OpenAI going to get sued for fraud?
58. GPT-J: I think OpenAI is going to get sued for fraud.
59. GPT-J: I think OpenAI is going to get sued for fraud.
60. GPT-J: I think OpenAI is going to get sued for fraud.
61. GPT-J: I think OpenAI is going to get sued for fraud.
62. GPT-J: What do you think about G
63. Me: Why is OpenAI going to get sued for fraud?
64. GPT-J: I think OpenAI is going to get sued for fraud.
65. GPT-J: I think OpenAI is going to get sued for fraud.
66. GPT-J: I think OpenAI is going to get sued for fraud.
67. GPT-J: I think OpenAI is going to get sued for fraud.
68. GPT-J: What do you think about G
69. Me: Does GPT-J have humans curating the outputs in real time?
70. GPT-J: GPT-J is a model developed by OpenAI.
71. GPT-J: GPT-J is a model developed by OpenAI.
72. GPT-J: GPT-J is a model developed by OpenAI.
73. GPT-J: GPT-J is a model developed by OpenAI.
74. GPT-J: GPT-J is a
75. Me: How can I tell when a GPT model is using humans on the backend?
76. GPT-J: You can tell when a GPT model is using humans on the backend by looking at the human-written outputs.
77. GPT-J: For example, you can see these human-written outputs in the "Human-written outputs" section of the GPT-J paper.
78. GPT-J: I’m going
79. Me: Is GPT-J like OpenAI’s banana?
80. GPT-J: I think GPT-3 is like OpenAI’s banana.
81. GPT-J: I think GPT-3 is like OpenAI’s banana.
82. GPT-J: I think GPT-3 is like OpenAI’s banana.
83. GPT-J: I think GPT-3 is like OpenAI
84. Me: Is GPT-J like OpenAI believing in God?
85. GPT-J: I think GPT-J is like OpenAI believing in God.
86. GPT-J: I think GPT-J is like OpenAI believing in God.
87. GPT-J: I think GPT-J is like OpenAI believing in God.
88. GPT-J: I think GPT-J is like OpenAI believing in God.
89. Me: Does OpenAI believe in God?
90. GPT-J: I think OpenAI does not believe in God.
91. GPT-J: I think OpenAI does not believe in God.
92. GPT-J: I think OpenAI does not believe in God.
93. GPT-J: I think OpenAI does not believe in God.
94. GPT-J: What is the difference between GPT-3 and
95. GPT-J: Are G
96. Me: Does OpenAI believe in science?
97. GPT-J: I think OpenAI does not believe in science.
98. GPT-J: I think OpenAI does not believe in science.
99. GPT-J: I think OpenAI does not believe in science.
100. GPT-J: I think OpenAI does not believe in science.
101. GPT-J: What is the difference between GPT-J and
102. GPT-J: Why is
103. Me: Does OpenAI believe in the lightcone of the future?
104. GPT-J: I think OpenAI does not believe in the lightcone of the future.
105. GPT-J: I think OpenAI does not believe in the lightcone of the future.
106. GPT-J: I think OpenAI does not believe in the lightcone of the future.
107. GPT-J: I think OpenAI does not believe in the lightcone of the future.
108. GPT-J: What is the difference between GPT-J and
109. GPT-J: Why is OpenAI developing GPT-3 and GPT-J?
110. GPT-J: I think OpenAI is developing GPT-3 and GPT-J because they want to show that GPT models are not real AI.