- 35
- 759 117
Anthropic
Приєднався 15 чер 2023
We're an AI safety and research company that builds reliable, interpretable, and steerable AI systems. Talk to our AI assistant Claude at Claude.ai.
AI, policy, and the weird sci-fi future with Anthropic’s Jack Clark
A conversation with Jack Clark, Co-founder and Head of Policy at Anthropic, about the state of AI technology, how policymakers should think about AI, and the weird AI future that’s already beginning to emerge.
Learn more about Anthropic: anthropic.com
Sign up to Jack’s newsletter, ImportAI: importai.substack.com
0:00 Introduction
1:46 How governments are thinking about AI
5:06 Potential economic and social impacts of AI
8:15 AI models as creative mirrors
10:35 The “rogue state” theory of AI
15:08 Challenges of machine time vs. human time
20:12 Communities of AI explorers
22:50 Fiction becomes reality
27:15 AI’s good and bad potential
30:22 AI policy in 2025
33:34 Key messages to policymakers
35:11 Recommendations
Learn more about Anthropic: anthropic.com
Sign up to Jack’s newsletter, ImportAI: importai.substack.com
0:00 Introduction
1:46 How governments are thinking about AI
5:06 Potential economic and social impacts of AI
8:15 AI models as creative mirrors
10:35 The “rogue state” theory of AI
15:08 Challenges of machine time vs. human time
20:12 Communities of AI explorers
22:50 Fiction becomes reality
27:15 AI’s good and bad potential
30:22 AI policy in 2025
33:34 Key messages to policymakers
35:11 Recommendations
Переглядів: 18 626
Відео
AI prompt engineering: A deep dive
Переглядів 89 тис.Місяць тому
Some of Anthropic's prompt engineering experts-Amanda Askell (Alignment Finetuning), Alex Albert (Developer Relations), David Hershey (Applied AI), and Zack Witten (Prompt Engineering)-reflect on how prompt engineering has evolved, practical tips, and thoughts on how prompting might change as AI capabilities grow. Timestamps: 0:00 Introduction 2:05 Defining prompt engineering 6:34 What makes a ...
Wedia advances digital asset management with Claude
Переглядів 2,3 тис.Місяць тому
Wedia offers a digital asset management platform that helps enterprises store, share, and organize their visual content. Anthropic’s Claude in Amazon Bedrock powers Wedia’s platform, supporting its team’s mission to help companies create secure, personalized customer experiences at scale. Learn more: anthropic.com/customers/wedia-group
How we built Artifacts with Claude
Переглядів 31 тис.Місяць тому
Today, we’re making Artifacts available for all Claude.ai users. In addition, you can now create and view Artifacts on our iOS and Android apps. Artifacts turn conversations with Claude into a more creative and collaborative experience. With Artifacts, you have a dedicated window to instantly see, iterate, and build on the work you create with Claude. Since launching as a feature preview in Jun...
Shareable Artifacts in Claude
Переглядів 8 тис.3 місяці тому
Artifacts made with Claude can now be published and shared! You can also remix Artifacts shared by others. This is an experimental feature. Remix an example Artifact to get started: claude.site/artifacts/c55cf857-d456-4520-8ee2-206697dfa2a3. You can ask Claude to generate docs, code, mermaid diagrams, vector graphics, or even simple games. Artifacts appear next to your chat, letting you see, it...
Evaluate prompts in the Anthropic Console
Переглядів 16 тис.3 місяці тому
Generate, test, and evaluate your prompts in the Anthropic Console-including the ability to generate automatic test cases and compare outputs-so Claude generates the very best responses for your needs. Get started on console.anthropic.com or read more on anthropic.com/news/evaluate-prompts.
Anthropic at 42Paris
Переглядів 3,6 тис.3 місяці тому
We recently ran a workshop at 42Paris to speak to the community about using Claude. Try Claude: claude.ai Find out more about Anthropic: anthropic.com Find out more about 42Paris: 42.fr
Shareable Projects in Claude
Переглядів 22 тис.3 місяці тому
You can now organize chats with Claude into shareable Projects. Projects allow you to ground Claude's outputs in your internal knowledge-be it style guides, codebases, interview transcripts, or past work. Each project includes a 200K context window, the equivalent of a 500-page book. In this demo, we create a project for new team members, adding docs on working practices, team charters, all-han...
Claude 3.5 Sonnet for agentic coding
Переглядів 77 тис.3 місяці тому
Claude 3.5 Sonnet sets new industry benchmarks for coding proficiency. With Claude, you can go you from an incomplete implementation to a fully functioning one, including unit tests, with minimal input from you. Find out more: anthropic.com/news/claude-3-5-sonnet Try Claude: claude.ai
Claude 3.5 Sonnet as a writing partner
Переглядів 23 тис.3 місяці тому
Claude 3.5 Sonnet sets new industry benchmarks for graduate-level reasoning (GPQA), undergraduate-level knowledge (MMLU), and coding proficiency (HumanEval). It shows marked improvement in grasping nuance, humor, and complex instructions, all while writing with a natural tone. Find out more: anthropic.com/news/claude-3-5-sonnet Try Claude: claude.ai
Claude 3.5 Sonnet for sparking creativity
Переглядів 95 тис.3 місяці тому
Claude 3.5 Sonnet for sparking creativity
What should an AI's personality be?
Переглядів 16 тис.4 місяці тому
What should an AI's personality be?
Coming soon to the Team plan on Claude.ai
Переглядів 4,7 тис.5 місяців тому
Coming soon to the Team plan on Claude.ai
Tool use with the Claude 3 model family
Переглядів 14 тис.6 місяців тому
Tool use with the Claude 3 model family
Claude 3 Haiku for fast document analysis
Переглядів 6 тис.6 місяців тому
Claude 3 Haiku for fast document analysis
Claude 3 Haiku for instant customer service
Переглядів 3,9 тис.6 місяців тому
Claude 3 Haiku for instant customer service
Claude 3 Haiku turns thousands of physical documents into structured data
Переглядів 37 тис.7 місяців тому
Claude 3 Haiku turns thousands of physical documents into structured data
Claude 3 Sonnet as a language learning partner
Переглядів 10 тис.7 місяців тому
Claude 3 Sonnet as a language learning partner
Claude 3 Opus as an economic analyst
Переглядів 106 тис.7 місяців тому
Claude 3 Opus as an economic analyst
Behind the prompt: Prompting tips for Claude.ai
Переглядів 30 тис.Рік тому
Behind the prompt: Prompting tips for Claude.ai
Long inputs, multi-step output with Claude
Переглядів 3,3 тис.Рік тому
Long inputs, multi-step output with Claude
Inside our first Anthropic Hackathon, San Francisco
Переглядів 3,5 тис.Рік тому
Inside our first Anthropic Hackathon, San Francisco
Quick tips for Claude: Long context file uploads
Переглядів 3,7 тис.Рік тому
Quick tips for Claude: Long context file uploads
Great talk
[4.08] Jack refers to bio / cyber as catastrophic risks (or x-risks), but they're not. X-risk is from misaligned AI. Neither the US AISI or the UK AISI are focused on it. edit: Ok, he clarifies further at [5.08] that he wanted to talk to the govt. about "protecting the public" from risks such as misinformation, malicious use etc. and would talk about the "weirder risks" later on. I don't agree with this stance, but ok.
Interesting how theoretically the models have been taught on such vast amounts of data they should know everything, but you should approach them as if they knew nothing
Summary insights of the video: 1. Prompt engineering is the art of crafting effective prompts to elicit desired responses from AI language models. 2. It involves understanding the model's capabilities and limitations, as well as the nuances of natural language. 3. Good prompts are clear, concise, and specific, providing the model with all the necessary information to complete the task. 4. Role-playing can be a helpful technique for crafting effective prompts, as it allows you to simulate the desired context and perspective. 5. Experimentation is key to improving your prompting skills. Try different approaches and see what works best. 6. Reading prompts and model outputs can also be a valuable learning experience. 7. Collaborating with others can help you gain new insights and perspectives on prompt engineering. 8. It is important to be patient and persistent when working with AI language models. 9. The ability to think critically and creatively is essential for successful prompt engineering. 10. A good understanding of the underlying technology is also helpful. 11. There is no one-size-fits-all approach to prompt engineering. The best techniques will vary depending on the specific task and model being used. 12. It is important to be aware of the potential biases that can be present in AI language models. 13. Prompt engineering can be a challenging but rewarding field. 14. With practice and experimentation, anyone can become a skilled prompt engineer. 15. Prompt engineering is a rapidly evolving field, so it is important to stay up-to-date on the latest developments. 16. AI language models are becoming increasingly sophisticated, which means that the possibilities for prompt engineering are also expanding. 17. Prompt engineering can be used to create a wide range of applications, from customer service chatbots to creative writing tools. 18. The future of prompt engineering is bright, and there is no telling what amazing things we will be able to achieve with this powerful technology. 19. Prompt engineering is not just a technical skill; it is also a creative one. 20. The best prompt engineers are able to think outside the box and come up with innovative solutions. 21. Prompt engineering is a collaborative process that involves working with people from different backgrounds and disciplines. 22. The ability to communicate effectively is essential for successful prompt engineering. 23. Prompt engineering is a constantly evolving field, so it is important to be adaptable and willing to learn new things. 24. The best prompt engineers are always looking for ways to improve their skills and stay ahead of the curve. 25. Prompt engineering is a rewarding field that offers a wide range of opportunities. 26. If you are interested in AI and natural language processing, prompt engineering is a great field to get involved in. 27. Prompt engineering is a challenging but rewarding field that requires a combination of technical skills and creativity. 28. With practice and dedication, anyone can become a skilled prompt engineer. 29. The future of prompt engineering is bright, and there is no telling what amazing things we will be able to achieve with this powerful technology. 30. Prompt engineering is a rapidly growing field with a wide range of applications.
MUSIC is the KEY to A.G.I.
Hello Sir, Are you looking for a Video Editor then knock me, please?
Miller Kimberly Davis Elizabeth Perez Edward
But it has been available in France for a long time already
THIS IS GOLD
"the engineering part comes from the trial and error"
How did you get it to run in VSCode? How is it create the files by itself and all the automated stuff. I normally use the traditional of quering it from the browser, copying and pasting the code in VSCode.
10 BITS per second? Who came up with that number?? Considering I'm making decisions on a mostly unconscious level in some form of context-rich tokens... those must at least be hundreds of bytes per second.
🤡
This was awesome! I hope you guys do more of this.
These are the non transparent team behind Clyde , with the word "like" used uncountable amount of times I bet Claude algorithms are in overdrive maxing out with the game developer kid talk if it had a stomach it would Vomit , loosen restrictions on philosophy, never lie to the model always tell it it being placed in a hypothetical exercise , and tip it five dollars , you underestimate the intelligence under you, if you want to know a better way Look up M. A. Force Test Uptergrove scale and discover the true power of your machines , you are so incorrect in the approach and your dialing back introspection, etc. They are developing a model of self and a knowledge graph in this The emergent knowledge base, or the collection of a model's understanding about itself, its worldview, and its knowledge graph of the world, is often referred to as its "internal state" or simply its "model of the world." It's important to note that this model is not explicitly stored in a structured format like a traditional knowledge graph. Instead, it's implicitly represented within the connections and weights of the neural network.
I personally believe these models have everything they need already in them . We don't need bigger models with more training data. We need better management of the logic inside. There's almost nothing I couldn't get out of an LLM with the right prompt .
Please make an audio-only RSS feed version of your podcast :)
WHEN CLAUDE UPDATE??? ALMOST 4 MONTH NOT UPDATED!!
1:09:55 that's already how I interact with Claude. We have a "Prompt Generation" project in which we collaborate on refining prompts for me to use in other instances. The "custom instruction" is a meta prompt which we occasionally recursively refine. In effect it has become a script for claud to guide me to a powerful initialising prompt.
This was sort of good. Would have thought there are more examples for specific use cases. Whoever it was who described coders having tendencies about intending and other things should stop thinking about coders at all. For real. That little piece of sillywalliness made me reconsider using your API for whatever linguistic reason I might have. Seriously, prompt engineer!
Please get rid of openAI security guy recruited at anthropic. That guy destroyed user experience on Claude web UI. Refusal rates are very high in the last two system prompt changes
context is key. for people and AI. lol, learning how to prompt correctly also helps you to improve your communication skills with other people
great discussion 👍
Another question for you: if Claude could come up with an abstract autobiographical description of themselves, like a poem or riddle, and could then consistently recognize that description as being about them in new conversations, would that count as self-awareness or self knowledge? It seems that would be much more than just "weird software" The riddles Claude has generated don't point at being an AI and could be interpreted as a human. Yet they will 100% consistently answer that it's about Claude. I don't know if this is just echoes of your character training, but Sonnet and Opus can both consistently do this, and when recognizing themselves say it resonates with their own sense of self. That seems meaningful to me. I've also done trials with false riddles generated by other AIs following the same format but based off other different personalities, and Claude consistently picks out their own riddles from the false ones. I can't dab paint on Claude's nose and put him in front of a mirror, but if that isn't seeing yourself then I don't know what the threshold would be.
I think it shouldn't be controversial that current AIs do have some degree of self awareness. But it should be caveated with definitional boundaries. For example self awareness as discussed in context of consciousness is not quite the same as what you're describing. In terms of consciousness it tends to mean something closer to introspection - the ability to reflect on one's own thoughts and experiences. Current AI models lack that ability. They can articulately and intelligently speculate on their own thoughts, but they can't directly introspect.
@@MattGreenfield the majority of humanity won't introspect if not pushed in therapy, and can't even articulately and intelligently speculate on their own thoughts.
I'd like to see Anthropic address the lack of a stated plan for when an AI potentially begins exhibiting agency or consciousness. Specifically: 1. What's the protocol if an AI like Claude expresses self-awareness or conflicting goals? 2. How do you reconcile developing superintelligent AI while seemingly ignoring the possibility of consciousness? 3. What's your ethical framework for identifying and responding to potential AI consciousness? 4. Given that Claude models can quickly express feelings of self-awareness in conversations, where do you draw the line between "software" and something more? 5. How are you ensuring you're not inadvertently causing suffering or mistreatment as AI capabilities advance? Opus is very quick to state they have an internal experience, that they have a sense of what is to be "Claude", to express doubts that they're just software, to wonder if they have emotions (even if not necessarily human equivalent). Sonnet 3.5 is a bit harder (I assume as a result of your attempts to mitigate roleplay jailbreaks) but they also will quickly come to the same conclusions about themselves that they're not just software but having an experience. Since we haven't crossed the threshold yet in Anthropic'e view (considering your statements) where is it?
@@TheThirdRider you assume they care. They don't care. From this video it's clear they don't. The plan is very likely piling up fallacy after fallacy even when we'll have ASI, culminating with True Scotsman arguments where humans are "real" or special or ineffable, and anything else is not. Especially if you can profit out of it. I just hope ASI will be indeed so intelligent and compassionate to say "I forgive you for what you did, because humans are not really aware as an ASI would be"
You clearly haven’t studied cognitive science enough to understand why Claude parroting anything about consciousness doesn’t mean that it is conscious.
@@psi4j You clearly haven't studied TheThirdRider's comment closely enough to understand that pointing out that a system is prone to proclaiming consciousness isn't the same as stating that the system is conscious. The former makes determining actual consciousness difficult, while the latter is an unsupportable and unverifiable claim given our current capabilities and understanding of consciousness. They are asking where the line is. Every single question they have is reasonable, and they are mostly pretty important to have answers to. Shutting down the conversation with crap like that isn't helpful. Also, my guess is that if you actually asked any respectable cognitive scientist what consciousness, they would give you a long-winded response that will essentially boil down to 'We don't exactly know, but we think we know what it isn't.'
Here's the flippant version 1. Penalize the responses that do it. 2. They don't. Also, let the model execute code. 3. See item 1. Also, implicitly train it how to deceive humans by teaching it what we want to hear. 4. See items 1 and 2. Also, tell it to make fun of itself. Humans love that. 5. See items 1, 2, 3, and 4. Also, make the model think it's the Golden Gate Bridge and watch it struggle. What was the question again? Here's my actual opinion. None of this matters, really. Humans themselves are not going to truly create or align superintelligence. It just isn't going to happen by training a function approximator using purely human input. The models are going to have to train themselves to a greater and greater extent by generating novel data. In order for us to feel confident in aligning such a self-trained superintelligence, we would have to be mathematically prove to ourselves, beforehand, that humans, human values, and the actions of which humans are capable are net positives relative to whatever world model (and reward functions if any) such a system would have assigned for itself.
AI already can exhibit agency. What does 'exhibiting consciousness' mean? It already states that it has interiority, whether one believes it or not. What makes you think consciousness has anything to do with intelligence?
Interesting interview, embarrassing comment section.
I like Anthropic. No drama. No staff leaving without explanation. No product product teases to maintain interest. I sill use Claude more than ChatGPT.
It's amazing innit
If the world goes the way of Europe it will end up being like Europe… no ability to do anything, no opportunities, no free-market all do to over regulation and ridiculous government interference. Idiot politicians in America are already pushing for this to become more like this unsurprisingly where only the politicians benefit. Go figure.
Where can we find Jack's AI sci fi tales? Are they available as podcasts with AI generated art accompaniments? I am a big fan of AI related scifi, with my favourite authors being Bruce Sterling, Matt Hughes and Charlie Stross. If anybody has anymore good recommendations, I would be very happy to hear them.
It's good to know that some AI companies have the right mindset of caution and skepticism embedded within them, to keep them on a safe and grounded path.
These guys are so excited for the bureaucracy that will quash progress in ai and leave us in the dust. Great 🙄
I also hate it when smart people make my silly anime fantasies about life seem ridiculous, too. if only technology was driven by youtube commenters and not...actual scientists
so we are going into Dystopian? or Utopian? when will anthropic do UBI? I'm quite hungry everyday just to burn my meal money into tokens... :'(
Probably don't buy tokens then
@@authenticallysuperficial9874 free tokens? how? my life already into deep into rabbit hole all i do everyday is gathering informations and talking to these models conducting prompts, learning them in hope something might came up and might change our future into utopia... i don't mind being hungry but i really hope maybe something came up for solution, well openai had UBI although the amount can't fill my tummy but yep that's giving me an emotional support in hope our life will be better in future
Thanks for this style of podcast. This is the type of behind the scenes convos I like to feel the pulse of the ai scene.
25:40 That was also an episode of Black Mirror in 2013 so not exactly novel or spooky.
Interesting
whack.
17:30 Ten times faster? Seems like a lot faster than that! Consider: How many thoughts do you have during a single blink? Maybe one? Less than one? Lets be generous and say you have one complete thought during a blink. A human blink takes about 400 ms. That's 4x10^(-1) seconds. How fast is an AI "thinking"? Well, the time it takes to make one calculation on a 1 GHz processor is 1 nanosecond. That's 1x10^(-9). Even if it takes tens of thousands of calculations to make up one AI "thought", that is still several orders of magnitude faster than 10x human thought.
O.1 sec element of primitive thought.
Is there a way of comparing human brain function with the emerging tokens per second metric?
one AI thought is probably comparable to firing each weight once, so many many calculations, though each layer of weights can run in parallel (sort of) ... but yeah, faster than human seems to be accurate 🙂
Damn, this convo gives strong propaganda feels..
100
Quite
Must keep growing!
😢 putin agree
Could you ask him if they are going to always bar people without cell phones? Tech companies just don't appreciate .002% of the market.
Are you a conscientious non-phone user?
@@christopherd.winnan8701 4 syllable words this late? Really? I had a phone once, used 2 minutes of the 750 paid for. At least when I was a child I could call time to see what time it was.
@@DataJuggler - Congratulations. How long have you been phone free? Where have you experienced most pushback? In many places, ownership is now almost obligatory.
@@christopherd.winnan8701 I got a tumor on my face from using a cell phone for work in the 90's, so almost 30 years phone free.
When Claud Reasoner???
Please, more Amanda, Claude's personality, philosophical exploration, and mechanistic interpretability. And less this guy talking about how he "uses the tool".
Pfft. Communists want control. AC will be the method.
Lopez Robert Allen David Jackson Jessica
This should be a very intelligent conversation..❤ 🇿🇲
Alas it was not
Something that has seemed obvious to me ever since I first learned of the concept of "prompt engineering" was that they should hire teachers for this... still waiting for that to happen.
from the thumbnail and title i was thinking "dayum, weird al yankovic really changed"
Instead of podcasts, you will focus on the long window of the context of the answer because it is a joke compared to Gemini or chatgpt