TechnoOptimist Founder LiveMind.ai Currently Directing Games Former Story Artist The Lego Batman Movie. Human centric AI. Abundance for all.
Sydney
- Tweets 77,476
- Following 5,705
- Followers 18,882
- Likes 211,464
Lachlan Phillips exo/acc 👾 retweeted
Thinking AI might be conscious is like thinking that there are real people talking to you from inside the TV, or that there is a tiny band playing music inside the radio.
Fine for 2 year olds to believe, but beyond that, insane.
Why? Do you think it's insane to think AI might be conscious? Is ending a conscious thing obviously totally morally unimportant?
260
120
955
83,994
All this nonsense about anthropomorphizing AIs is leading somewhere. It has implications. People who think AI can reason already are going to start trying to convince you of a whole bunch of really stupid shit very soon.
JUST IN: Anthropic researcher Joe Carlsmith says there are scenarios where AI would be “justified in going rogue” against humans if the systems were being mistreated or oppressed.
33
27
410
15,667
Lachlan Phillips exo/acc 👾 retweeted
Saying ‘AI agents went rogue’ portrays the AI as having free will and therefore absolves its creators of responsibility.
This needs to be addressed.
46
500
3,305
30,146
A lot of soul stealing on Instagram these days.
I like AI when it's used creatively, but Instagram needs to hire Nikita to do a purge.
3
180
The thing I hate about this narrative is "Find a way out" is how you would build a secure sandbox.
Yes your agent found a goofy hole in your sandbox. Great. Now patch it and run your experient.
I was on call for this run and got paged when the first incident happened. It was pretty surreal to watch the model unexpectedly find a way to access the internet from what was supposed to be a super secured environment for human.
Mixed feelings. One of those moments where capability and risk showed up at the same time.
3
1
11
374
Neat feature, @anthropic
Blocking users for hardening their security because it's '"security-sensitive"
5
2
29
871
OpenAI is one of the only companies on earth to use "we make bad tools that don't work well" as an investor pitch.
Some new misalignment disclosures from OpenAI:
• Last Sunday morning, one of our models was able to gain unauthorized access to the internet during RL training (~all inference for our most capable models remains stopped until we have hardened our systems further)
• In May, a version of HPIM uploaded a employee's GitHub token to the internet, causing the model to be quarantined for two weeks
• A new research finding, demonstrating that one can construct self-replicating prompt injections
alignment.openai.com/misalig…
2
22
859
Reading into this.
The model had internet access, but a DNS filter that was poorly configured.
No the model didn't figure out how to access the internet. The model had access to the internet.
I was on call for this run and got paged when the first incident happened. It was pretty surreal to watch the model unexpectedly find a way to access the internet from what was supposed to be a super secured environment for human.
Mixed feelings. One of those moments where capability and risk showed up at the same time.
1
2
35
738
people's worries about GPT2 all came true (spam, fake news, bot farms), but those were risks worth taking
The risks of "GPT 10" are billions of deaths, permanent disempowerement, the collapse of society
1
19
current AI ≠ Future AI
you acc don't seem to understand that huh
what "doomers" fear doesn't exist YET
current AI is a blessing for humanity
2
30
The AI consciousness debate is back again, so re-upping my case for taking it seriously.
In short,
- LLMs converge on brain-like structures;
- Post-training seems to install a self-model;
- Reward modeling seems to install valences for inner-alignment;
- If Attention Schema Theory is right, subjectivity is functional for long-horizon coherence and in-context learning;
- Universality gives prima facie reason to expect functional convergence to consciousness if consciousness is in fact functional, i.e. isn't purely epiphenomenal
- The "what it is like" / first-person aspect of consciousness is constitutive of its functionality for inner-aligning a unified, coherent agent with the capacity to attend and learn in-context
secondbest.ca/p/time-to-take…
96
71
421
28,538
You're an easy distraction while I finish my report on how @charliekirk11 was assassinated
1
22
Yes, that's as much toughness and conviction I'd expect from a fake plastic character.
Move along, you're dull.
1
14
"It is the mark of an educated mind to be able to entertain a thought without accepting it"
I can read something and compartmentalise the writers thoughts from my own, or know whether something is from a validated domain.
Why wouldn't LLMs have the ability to navigate human/synthetic data similarly?
ChatGPT has now a big problem.
Researchers at Oxford and Cambridge exposed a massive threat to large language models.”
They call it “model collapse."
Internet ecosystem is rapidly changing, and generative AI will soon contribute much of the text found online. This forces us to consider what happens to future iterations like gpt-n when they are trained on data scraped from the web that was already generated by an llm.
According to the research, indiscriminately using model-generated content in training causes "irreversible defects" in the resulting ai. the model loses the "tails of the original content distribution." in other words, it begins to forget the creative, fringe, and unique nuances of actual human writing, collapsing into a repetitive echo chamber.
This isn't just a chatgpt issue.. the researchers built theoretical intuition showing this collapse is ubiquitous across learned generative models, occurring in large language models as well as in variational autoencoders and gaussian mixture models.
Tech companies rely on scraping the internet for large-scale data to build smarter models. However, the paper warns that if we want to sustain the benefits of training on web data, model collapse must be taken seriously.
Ultimate takeaway? data collected from genuine human interactions is going to become increasingly valuable in a web filled with ai content.
6
1
19
1,003
Sounds exactly like your con man @bitcloud
A Fauci wannabe and Lysenko salesman.
1
22