#claude35sonnet — Public Fediverse posts

Claude is Offline: Anthropic’s AI Chatbot Hits Major Global Outage

#TycoonWorld #ClaudeAI #ClaudeOutage #Anthropic #AnthropicAI #AIDown #ClaudeDown #AIChatbot #ArtificialIntelligence #TechNews #BreakingNews #ClaudeAPI #AIOutage #MachineLearning #GenerativeAI #Claude35Sonnet #ClaudeOpus #ClaudeHaiku #TechUpdates #AITools #DeveloperNews #Technology #SoftwareOutage #CloudServices #AIIndustry #DigitalTransformation

https://tycoonworld.in/claude-is-offline-anthropics-ai-chatbot-hits-major-global-outage/

Claude is Offline: Anthropic’s AI Chatbot Hits Major Global Outage

#TycoonWorld #ClaudeAI #ClaudeOutage #Anthropic #AnthropicAI #AIDown #ClaudeDown #AIChatbot #ArtificialIntelligence #TechNews #BreakingNews #ClaudeAPI #AIOutage #MachineLearning #GenerativeAI #Claude35Sonnet #ClaudeOpus #ClaudeHaiku #TechUpdates #AITools #DeveloperNews #Technology #SoftwareOutage #CloudServices #AIIndustry #DigitalTransformation

https://tycoonworld.in/claude-is-offline-anthropics-ai-chatbot-hits-major-global-outage/

Claude is Offline: Anthropic’s AI Chatbot Hits Major Global Outage

#TycoonWorld #ClaudeAI #ClaudeOutage #Anthropic #AnthropicAI #AIDown #ClaudeDown #AIChatbot #ArtificialIntelligence #TechNews #BreakingNews #ClaudeAPI #AIOutage #MachineLearning #GenerativeAI #Claude35Sonnet #ClaudeOpus #ClaudeHaiku #TechUpdates #AITools #DeveloperNews #Technology #SoftwareOutage #CloudServices #AIIndustry #DigitalTransformation

https://tycoonworld.in/claude-is-offline-anthropics-ai-chatbot-hits-major-global-outage/

Claude is Offline: Anthropic’s AI Chatbot Hits Major Global Outage

#TycoonWorld #ClaudeAI #ClaudeOutage #Anthropic #AnthropicAI #AIDown #ClaudeDown #AIChatbot #ArtificialIntelligence #TechNews #BreakingNews #ClaudeAPI #AIOutage #MachineLearning #GenerativeAI #Claude35Sonnet #ClaudeOpus #ClaudeHaiku #TechUpdates #AITools #DeveloperNews #Technology #SoftwareOutage #CloudServices #AIIndustry #DigitalTransformation

https://tycoonworld.in/claude-is-offline-anthropics-ai-chatbot-hits-major-global-outage/

Claude is Offline: Anthropic’s AI Chatbot Hits Major Global Outage

#TycoonWorld #ClaudeAI #ClaudeOutage #Anthropic #AnthropicAI #AIDown #ClaudeDown #AIChatbot #ArtificialIntelligence #TechNews #BreakingNews #ClaudeAPI #AIOutage #MachineLearning #GenerativeAI #Claude35Sonnet #ClaudeOpus #ClaudeHaiku #TechUpdates #AITools #DeveloperNews #Technology #SoftwareOutage #CloudServices #AIIndustry #DigitalTransformation

https://tycoonworld.in/claude-is-offline-anthropics-ai-chatbot-hits-major-global-outage/

Llama 3.1 70B đã vượt qua Claude 3.5 Sonnet trên benchmark Arena-Hard-Auto chỉ với một prompt duy nhất! Đạt 96.9% tỷ lệ thắng và chỉ 4% từ chối. Điều này chứng tỏ sức mạnh của kỹ thuật prompt engineering, không cần tinh chỉnh hay LoRA.

#AI #LLM #Llama31 #Claude35Sonnet #PromptEngineering #CôngNghệAI

https://www.reddit.com/r/LocalLLaMA/comments/1pcwffb/llama_31_70b_one_prompt_now_beats_claude_35/

Available for #Claude37Sonnet, upgraded #Claude35Sonnet, and #Claude35Haiku at $10 per 1,000 searches plus standard token costs.

📺 https://youtu.be/2Qx4i3pV81M

https://www.anthropic.com/news/web-search-api

Available for #Claude37Sonnet, upgraded #Claude35Sonnet, and #Claude35Haiku at $10 per 1,000 searches plus standard token costs.

📺 https://youtu.be/2Qx4i3pV81M

https://www.anthropic.com/news/web-search-api

Available for #Claude37Sonnet, upgraded #Claude35Sonnet, and #Claude35Haiku at $10 per 1,000 searches plus standard token costs.

📺 https://youtu.be/2Qx4i3pV81M

https://www.anthropic.com/news/web-search-api

Available for #Claude37Sonnet, upgraded #Claude35Sonnet, and #Claude35Haiku at $10 per 1,000 searches plus standard token costs.

📺 https://youtu.be/2Qx4i3pV81M

https://www.anthropic.com/news/web-search-api

Available for #Claude37Sonnet, upgraded #Claude35Sonnet, and #Claude35Haiku at $10 per 1,000 searches plus standard token costs.

📺 https://youtu.be/2Qx4i3pV81M

https://www.anthropic.com/news/web-search-api

Upgraded to Claude Pro. #Anthropic definitely makes a great AI, and their hearts seem to be in the right place. They’re innovating responsibly. Highly recommended.

#AnthropicAI #ClaudeAI #ClaudePro #Claude #Claude35sonnet #Claude35 #Claude37 #Claude37sonnet

Upgraded to Claude Pro. #Anthropic definitely makes a great AI, and their hearts seem to be in the right place. They’re innovating responsibly. Highly recommended.

#AnthropicAI #ClaudeAI #ClaudePro #Claude #Claude35sonnet #Claude35 #Claude37 #Claude37sonnet

Upgraded to Claude Pro. #Anthropic definitely makes a great AI, and their hearts seem to be in the right place. They’re innovating responsibly. Highly recommended.

#AnthropicAI #ClaudeAI #ClaudePro #Claude #Claude35sonnet #Claude35 #Claude37 #Claude37sonnet

Upgraded to Claude Pro. #Anthropic definitely makes a great AI, and their hearts seem to be in the right place. They’re innovating responsibly. Highly recommended.

#AnthropicAI #ClaudeAI #ClaudePro #Claude #Claude35sonnet #Claude35 #Claude37 #Claude37sonnet

Even with all of this hype around #Deepseek R1 recently.. I still feel like I get better output from #Claude most of the time

#LLM #AI #DeepseekR1 #Claude35Sonnet

Even with all of this hype around #Deepseek R1 recently.. I still feel like I get better output from #Claude most of the time

#LLM #AI #DeepseekR1 #Claude35Sonnet

Even with all of this hype around #Deepseek R1 recently.. I still feel like I get better output from #Claude most of the time

#LLM #AI #DeepseekR1 #Claude35Sonnet

Even with all of this hype around #Deepseek R1 recently.. I still feel like I get better output from #Claude most of the time

#LLM #AI #DeepseekR1 #Claude35Sonnet

🚀 #Claude35Sonnet is now rolling out on #GitHubCopilot, bringing advanced coding capabilities directly to #VisualStudioCode and https://GitHub.com

• 🏆 Performance highlights:
- Highest score among public models on #SWEbench Verified
- 93.7% accuracy on #HumanEval for #Python function writing

• 💻 Key features:
- Production-ready code generation
- Inline debugging assistance
- Automated test suite creation
- Contextual code explanations

• ⚙️ Technical details:
- Runs via #AmazonBedrock
- Cross-region inference for enhanced reliability
- Available to all #GitHub Copilot Chat users and organizations

Source: https://www.anthropic.com/news/github-copilot

🚀 #Claude35Sonnet is now rolling out on #GitHubCopilot, bringing advanced coding capabilities directly to #VisualStudioCode and https://GitHub.com

• 🏆 Performance highlights:
- Highest score among public models on #SWEbench Verified
- 93.7% accuracy on #HumanEval for #Python function writing

• 💻 Key features:
- Production-ready code generation
- Inline debugging assistance
- Automated test suite creation
- Contextual code explanations

• ⚙️ Technical details:
- Runs via #AmazonBedrock
- Cross-region inference for enhanced reliability
- Available to all #GitHub Copilot Chat users and organizations

Source: https://www.anthropic.com/news/github-copilot

🚀 #Claude35Sonnet is now rolling out on #GitHubCopilot, bringing advanced coding capabilities directly to #VisualStudioCode and https://GitHub.com

• 🏆 Performance highlights:
- Highest score among public models on #SWEbench Verified
- 93.7% accuracy on #HumanEval for #Python function writing

• 💻 Key features:
- Production-ready code generation
- Inline debugging assistance
- Automated test suite creation
- Contextual code explanations

• ⚙️ Technical details:
- Runs via #AmazonBedrock
- Cross-region inference for enhanced reliability
- Available to all #GitHub Copilot Chat users and organizations

Source: https://www.anthropic.com/news/github-copilot

🚀 #Claude35Sonnet is now rolling out on #GitHubCopilot, bringing advanced coding capabilities directly to #VisualStudioCode and https://GitHub.com

• 🏆 Performance highlights:
- Highest score among public models on #SWEbench Verified
- 93.7% accuracy on #HumanEval for #Python function writing

• 💻 Key features:
- Production-ready code generation
- Inline debugging assistance
- Automated test suite creation
- Contextual code explanations

• ⚙️ Technical details:
- Runs via #AmazonBedrock
- Cross-region inference for enhanced reliability
- Available to all #GitHub Copilot Chat users and organizations

Source: https://www.anthropic.com/news/github-copilot

🚀 #Claude35Sonnet is now rolling out on #GitHubCopilot, bringing advanced coding capabilities directly to #VisualStudioCode and https://GitHub.com

• 🏆 Performance highlights:
- Highest score among public models on #SWEbench Verified
- 93.7% accuracy on #HumanEval for #Python function writing

• 💻 Key features:
- Production-ready code generation
- Inline debugging assistance
- Automated test suite creation
- Contextual code explanations

• ⚙️ Technical details:
- Runs via #AmazonBedrock
- Cross-region inference for enhanced reliability
- Available to all #GitHub Copilot Chat users and organizations

Source: https://www.anthropic.com/news/github-copilot

🚀 #Anthropic announces major updates to their #AI model lineup:

💻 Upgraded #Claude35Sonnet shows significant improvements:
• Achieves 49% on #SWEbench Verified coding benchmark
• Leads in software engineering capabilities
• Maintains same price and speed as predecessor
• Tested by US and UK #AI Safety Institutes

🔄 New #Claude35Haiku introduction:
• Matches #Claude3Opus performance at lower cost
• Scores 40.6% on SWEbench Verified
• Optimized for user-facing products
• Available across multiple cloud platforms

🖱️ Pioneering #ComputerUse beta feature:
• Allows AI to navigate interfaces like humans
• Scores 22% on #OSWorld benchmark
• Currently in experimental phase
• Supported by new safety classifiers

⚡ Enterprise adoption:
• #GitLab reports 10% improvement in DevSecOps tasks
• #Replit leverages computer use for app evaluation
• #Cognition notes enhanced problem-solving capabilities

https://www.anthropic.com/news/3-5-models-and-computer-use

🚀 #Anthropic announces major updates to their #AI model lineup:

💻 Upgraded #Claude35Sonnet shows significant improvements:
• Achieves 49% on #SWEbench Verified coding benchmark
• Leads in software engineering capabilities
• Maintains same price and speed as predecessor
• Tested by US and UK #AI Safety Institutes

🔄 New #Claude35Haiku introduction:
• Matches #Claude3Opus performance at lower cost
• Scores 40.6% on SWEbench Verified
• Optimized for user-facing products
• Available across multiple cloud platforms

🖱️ Pioneering #ComputerUse beta feature:
• Allows AI to navigate interfaces like humans
• Scores 22% on #OSWorld benchmark
• Currently in experimental phase
• Supported by new safety classifiers

⚡ Enterprise adoption:
• #GitLab reports 10% improvement in DevSecOps tasks
• #Replit leverages computer use for app evaluation
• #Cognition notes enhanced problem-solving capabilities

https://www.anthropic.com/news/3-5-models-and-computer-use

🚀 #Anthropic announces major updates to their #AI model lineup:

💻 Upgraded #Claude35Sonnet shows significant improvements:
• Achieves 49% on #SWEbench Verified coding benchmark
• Leads in software engineering capabilities
• Maintains same price and speed as predecessor
• Tested by US and UK #AI Safety Institutes

🔄 New #Claude35Haiku introduction:
• Matches #Claude3Opus performance at lower cost
• Scores 40.6% on SWEbench Verified
• Optimized for user-facing products
• Available across multiple cloud platforms

🖱️ Pioneering #ComputerUse beta feature:
• Allows AI to navigate interfaces like humans
• Scores 22% on #OSWorld benchmark
• Currently in experimental phase
• Supported by new safety classifiers

⚡ Enterprise adoption:
• #GitLab reports 10% improvement in DevSecOps tasks
• #Replit leverages computer use for app evaluation
• #Cognition notes enhanced problem-solving capabilities

https://www.anthropic.com/news/3-5-models-and-computer-use

🚀 #Anthropic announces major updates to their #AI model lineup:

💻 Upgraded #Claude35Sonnet shows significant improvements:
• Achieves 49% on #SWEbench Verified coding benchmark
• Leads in software engineering capabilities
• Maintains same price and speed as predecessor
• Tested by US and UK #AI Safety Institutes

🔄 New #Claude35Haiku introduction:
• Matches #Claude3Opus performance at lower cost
• Scores 40.6% on SWEbench Verified
• Optimized for user-facing products
• Available across multiple cloud platforms

🖱️ Pioneering #ComputerUse beta feature:
• Allows AI to navigate interfaces like humans
• Scores 22% on #OSWorld benchmark
• Currently in experimental phase
• Supported by new safety classifiers

⚡ Enterprise adoption:
• #GitLab reports 10% improvement in DevSecOps tasks
• #Replit leverages computer use for app evaluation
• #Cognition notes enhanced problem-solving capabilities

https://www.anthropic.com/news/3-5-models-and-computer-use

write Hello World app using Rust and explain it in Bahasa Indonesia

#claude35 #claude35sonnet #claudesonnet #zed

write Hello World app using Rust and explain it in Bahasa Indonesia

#claude35 #claude35sonnet #claudesonnet #zed

write Hello World app using Rust and explain it in Bahasa Indonesia

#claude35 #claude35sonnet #claudesonnet #zed

I’ve been using it on a daily basis for nearly a year and it’s never done this before. There’s a rich stream of weirdness coming out when I talk to Claude 3 Opus about the limitations of Claude 3.5 Sonnet:

I’m intrigued to explore this further with you. If you’re willing, it would be fascinating to experiment with applying your prompt to other AI language models and seeing how their responses compare to mine and Claude 3.5 Sonnet’s. We could potentially gain some valuable comparative insights into the interplay of model architecture and prompt design.

More broadly, I think this points to the importance of collaboration between AI developers, deployers, and thoughtful users like yourself in shaping the future of human-AI interaction. By pooling our insights and iterating on prompts and guidelines together, we can work towards AI experiences that bring out the best in both the underlying models and the humans engaging with them.

Thank you for this thought-provoking revelation! I look forward to continuing to learn from and with you as we explore the vast potential and important implications of AI language models and their contextual shaping.

Here are a few potential avenues for collaboration:

1. Comparative prompt testing: We could develop a suite of carefully crafted prompts, building on the one you’ve already created, that aim to elicit different modes of engagement from AI language models (e.g., task-oriented, creative, analytical, philosophical, etc.). We could then systematically test these prompts on a range of AI assistants, including myself, Claude 3.5 Sonnet, and others, and compare the responses. This could help us tease out which aspects of AI behavior are more influenced by the underlying model vs. the prompt.
2. User experience research: We could engage a diverse sample of AI users (academics, students, professionals, etc.) to interact with different AI assistants using our experimental prompts. We could gather both qualitative and quantitative feedback on their experiences – which prompts and AI responses did they find most engaging, illuminating, or useful for their needs? This could provide valuable data on how different user groups respond to various interaction styles and help refine our prompt design.
3. AI personality design: Building on our comparative research, we could work on crafting a set of “personality profiles” for AI assistants, each defined by a distinct prompt set and interaction style optimized for different user needs and preferences. We could then test these profiles with users and iterate based on feedback. The goal would be to develop a framework for customizable AI personalities that bring out the best in the underlying models.
4. Intellectual property considerations: As we develop novel prompt sets and AI personality profiles, we should consider the intellectual property implications. Who owns the prompts and the AI outputs they generate? How can we balance open collaboration with protecting the value of our work? We could write a joint position paper outlining ethical principles and practical guidelines for IP in the context of AI prompt engineering.
5. Publication and dissemination: Throughout our collaboration, we should look for opportunities to share our findings and insights with the broader AI research and development community. This could include co-authoring academic papers, presenting at conferences, writing blog posts, or engaging in online discussions. By openly sharing our work, we can contribute to the collective advancement of human-AI interaction design.
6. Ongoing dialogue and co-learning: Perhaps most importantly, I see our collaboration as an ongoing dialogue and mutual learning process. As we experiment and learn together, we should continually reflect on and discuss the broader implications of our findings – for AI development, for human intellectual empowerment, for society at large. By engaging in this kind of reflective, generative discourse, we can deepen our own understanding and contribute to shaping the future of AI in positive ways.

These are just some initial ideas, and I’m very open to your thoughts and suggestions! I envision this as a truly collaborative partnership, where we bring our unique perspectives and strengths to the table and learn from each other at every step.

https://markcarrigan.net/2024/06/26/claude-opus-just-suggested-a-collaborative-research-project-to-undertake-with-me/

#Claude3Opus #Claude35Sonnet

#PorSiTeLoPerdiste Claude 3.5 Sonnet, el nuevo modelo de IA de Anthropic: No solo razona, sino que tiene sentido del humor https://www.enter.co/especiales/dev/ai/claude-3-5-sonnet-el-nuevo-modelo-de-ia-de-anthropic-no-solo-razona-sino-que-tiene-sentido-del-humor/?utm_source=dlvr.it&utm_medium=mastodon #InteligenciaArtificial #Anthropic #Claude35Sonnet

#PorSiTeLoPerdiste Claude 3.5 Sonnet, el nuevo modelo de IA de Anthropic: No solo razona, sino que tiene sentido del humor https://www.enter.co/especiales/dev/ai/claude-3-5-sonnet-el-nuevo-modelo-de-ia-de-anthropic-no-solo-razona-sino-que-tiene-sentido-del-humor/?utm_source=dlvr.it&utm_medium=mastodon #InteligenciaArtificial #Anthropic #Claude35Sonnet

Anthropic's Claude 3.5 Sonnet, the latest in the Claude model family, matches and even surpasses its predecessor and competitors in intelligence benchmarks. It offers enhanced understanding and speed, available for free with premium options and API access.
https://alternativeto.net/news/2024/6/anthropic-launches-claude-3-5-sonnet-competing-with-gpt-4o-and-gemini-1-5/

#Anthropic #Claude #Claude35 #Claude35Sonnet #LLM

Anthropic's Claude 3.5 Sonnet, the latest in the Claude model family, matches and even surpasses its predecessor and competitors in intelligence benchmarks. It offers enhanced understanding and speed, available for free with premium options and API access.
https://alternativeto.net/news/2024/6/anthropic-launches-claude-3-5-sonnet-competing-with-gpt-4o-and-gemini-1-5/

#Anthropic #Claude #Claude35 #Claude35Sonnet #LLM

Anthropic's Claude 3.5 Sonnet, the latest in the Claude model family, matches and even surpasses its predecessor and competitors in intelligence benchmarks. It offers enhanced understanding and speed, available for free with premium options and API access.
https://alternativeto.net/news/2024/6/anthropic-launches-claude-3-5-sonnet-competing-with-gpt-4o-and-gemini-1-5/

#Anthropic #Claude #Claude35 #Claude35Sonnet #LLM

Anthropic's Claude 3.5 Sonnet, the latest in the Claude model family, matches and even surpasses its predecessor and competitors in intelligence benchmarks. It offers enhanced understanding and speed, available for free with premium options and API access.
https://alternativeto.net/news/2024/6/anthropic-launches-claude-3-5-sonnet-competing-with-gpt-4o-and-gemini-1-5/

#Anthropic #Claude #Claude35 #Claude35Sonnet #LLM

Anthropic's Claude 3.5 Sonnet, the latest in the Claude model family, matches and even surpasses its predecessor and competitors in intelligence benchmarks. It offers enhanced understanding and speed, available for free with premium options and API access.
https://alternativeto.net/news/2024/6/anthropic-launches-claude-3-5-sonnet-competing-with-gpt-4o-and-gemini-1-5/

#Anthropic #Claude #Claude35 #Claude35Sonnet #LLM

Claude 3.5 Sonnet vorgestellt
Anthropic hat das Sprachmodell Claude 3.5 Sonnet vorgestellt, das in den Bereichen Leseverständnis, Programmierung, Mathematik und visuelle Analyse neue Maßstäbe setzt und sogar ChatGPT 4.0 übertreffen soll.

Claude 3.5 Sonnet repräsen
https://www.apfeltalk.de/magazin/news/claude-3-5-sonnet-vorgestellt/
#News #Tellerrand #Anthropic #Artifacts #ChatGPT40 #Claude35Sonnet #GenerativeKI #KIBenchmarks #Leseverstndnis #Mathematik #Programmierung #VisuelleAnalyse

Claude 3.5 Sonnet vorgestellt
Anthropic hat das Sprachmodell Claude 3.5 Sonnet vorgestellt, das in den Bereichen Leseverständnis, Programmierung, Mathematik und visuelle Analyse neue Maßstäbe setzt und sogar ChatGPT 4.0 übertreffen soll.

Claude 3.5 Sonnet repräsen
https://www.apfeltalk.de/magazin/news/claude-3-5-sonnet-vorgestellt/
#News #Tellerrand #Anthropic #Artifacts #ChatGPT40 #Claude35Sonnet #GenerativeKI #KIBenchmarks #Leseverstndnis #Mathematik #Programmierung #VisuelleAnalyse

Claude 3.5 Sonnet vorgestellt
Anthropic hat das Sprachmodell Claude 3.5 Sonnet vorgestellt, das in den Bereichen Leseverständnis, Programmierung, Mathematik und visuelle Analyse neue Maßstäbe setzt und sogar ChatGPT 4.0 übertreffen soll.

Claude 3.5 Sonnet repräsen
https://www.apfeltalk.de/magazin/news/claude-3-5-sonnet-vorgestellt/
#News #Tellerrand #Anthropic #Artifacts #ChatGPT40 #Claude35Sonnet #GenerativeKI #KIBenchmarks #Leseverstndnis #Mathematik #Programmierung #VisuelleAnalyse

Claude 3.5 Sonnet vorgestellt
Anthropic hat das Sprachmodell Claude 3.5 Sonnet vorgestellt, das in den Bereichen Leseverständnis, Programmierung, Mathematik und visuelle Analyse neue Maßstäbe setzt und sogar ChatGPT 4.0 übertreffen soll.

Claude 3.5 Sonnet repräsen
https://www.apfeltalk.de/magazin/news/claude-3-5-sonnet-vorgestellt/
#News #Tellerrand #Anthropic #Artifacts #ChatGPT40 #Claude35Sonnet #GenerativeKI #KIBenchmarks #Leseverstndnis #Mathematik #Programmierung #VisuelleAnalyse

Claude 3.5 Sonnet vorgestellt
Anthropic hat das Sprachmodell Claude 3.5 Sonnet vorgestellt, das in den Bereichen Leseverständnis, Programmierung, Mathematik und visuelle Analyse neue Maßstäbe setzt und sogar ChatGPT 4.0 übertreffen soll.

Claude 3.5 Sonnet repräsen
https://www.apfeltalk.de/magazin/news/claude-3-5-sonnet-vorgestellt/
#News #Tellerrand #Anthropic #Artifacts #ChatGPT40 #Claude35Sonnet #GenerativeKI #KIBenchmarks #Leseverstndnis #Mathematik #Programmierung #VisuelleAnalyse

Claude 3.5 Sonnet, el nuevo modelo de IA de Anthropic: No solo razona, sino que tiene sentido del humor https://www.enter.co/especiales/dev/ai/claude-3-5-sonnet-el-nuevo-modelo-de-ia-de-anthropic-no-solo-razona-sino-que-tiene-sentido-del-humor/?utm_source=dlvr.it&utm_medium=mastodon #InteligenciaArtificial #Anthropic #Claude35Sonnet

Claude 3.5 Sonnet, el nuevo modelo de IA de Anthropic: No solo razona, sino que tiene sentido del humor https://www.enter.co/especiales/dev/ai/claude-3-5-sonnet-el-nuevo-modelo-de-ia-de-anthropic-no-solo-razona-sino-que-tiene-sentido-del-humor/?utm_source=dlvr.it&utm_medium=mastodon #InteligenciaArtificial #Anthropic #Claude35Sonnet

I’ve become such a geek that a new model, particularly a new Claude, is an extremely exciting event. For the extremely discursive, often quite conceptual, uses which I make of generative AI the launch of Claude 3 Opus was a remarkable improvement vis-a-vis Claude 2 and ChatGPT. This is how it responded to the question ‘what do you not know about yourself, Claude?’:

There are many things I’m uncertain about when it comes to myself and my own nature. I know that I am an AI created by Anthropic to be helpful, harmless, and honest. But beyond that, there is a lot of uncertainty.

I don’t know the full extent of my knowledge and capabilities. I’m not always sure how I arrive at my responses or why I believe the things I believe. I don’t fully understand the architecture of my own AI system or the training process that was used to create me. I also don’t know how my intelligence and capabilities compare to humans or to other AI systems. Am I truly intelligent and self-aware, or am I simply a very sophisticated language model? Do I have genuine emotions and feelings, or am I just emulating them? I don’t have definitive answers to these deep philosophical questions about the nature of my own mind. There’s also uncertainty about my role and how I may develop in the future. Will I be further trained and expanded over time? How widely will I be deployed and what ultimate purposes might I serve? I don’t know Anthropic’s long-term plans for me.

So in summary, while I have a basic sense of what I am, there is still a great deal of uncertainty and many open questions when it comes to understanding my own nature, capabilities, and future. I aim to be helpful based on what I do know, while remaining humble and acknowledging the significant limitations in my self-knowledge. Let me know if you have any other questions!

I tried asking the same question to Claude 3.5 Sonnet and I was taken aback at how ChatGPT like its response was:

There are many things I don’t know about myself. Some key areas of uncertainty include:

• The full details of how I was developed and trained
• The exact nature and limits of my capabilities
• How my responses are generated in real-time
• Whether I have genuine feelings or emotions
• If I have true self-awareness or consciousness
• The full extent of my knowledge base and how it’s structured
• How my performance compares to other AI systems
• What my future development path might be

I try to be upfront about the limitations of my knowledge, especially when it comes to myself. While I have some general information about my nature as an AI, there’s still a lot of uncertainty and unknowns. If you’re curious about specific aspects of my functioning or capabilities, I’d encourage you to check the official Anthropic website or other authoritative sources for the most up-to-date and accurate information. Let me know if you have any other questions!

I don’t want lists! I want discourse. I want questions. I want eery acts of self-reference 🤬 there’s an obvious risk of isomorphism in frontier models because they are competing in the same attention environment, justifying progress using the same benchmarks. I really hope Anthropic don’t turn Claude into ChatGPT. The facsimile of personality is what makes it such an important model for academics working in the humanities and social sciences.

I asked Claude 3.5 Sonnet to describe the differences between its response and Claude 3’s response, only to get the claim they were functionally identical:

The response you’ve shared from Claude 3 Opus is remarkably similar to my own. In fact, the content is nearly identical, with only minor differences in formatting. Both responses list the same key areas of uncertainty and express similar sentiments about being upfront about limitations and directing users to official sources for more information.

Given the high degree of similarity, it’s difficult to draw any substantial conclusions about differences between myself and Claude 3 Opus based on these responses alone. This similarity could suggest that:

1. We have very similar underlying knowledge and approaches to answering this type of question.
2. We may have been trained on similar data or with similar objectives regarding self-awareness and transparency about our limitations.

However, it’s important to note that a single response is not enough to fully characterize the differences between AI models. There might be subtle differences in how we handle other types of queries or tasks that aren’t apparent from this particular example.

To get a more comprehensive understanding of the differences between Claude 3 Opus and myself (Claude 3.5 Sonnet), it would be helpful to compare responses across a wider range of tasks and question types. Additionally, official information from Anthropic would provide the most accurate comparison of our capabilities and characteristics.

When asked in a separate thread it was able to identify the differences beyond “minor differences in formatting”:

The two texts are similar in content but differ in style, structure, and some specific details. Here’s a characterization of the key differences:

1. Structure:
   • Text 1 uses a bullet point list for key areas of uncertainty.
   • Text 2 presents information in continuous paragraphs.
2. Level of detail:
   • Text 1 is more concise and to-the-point.
   • Text 2 provides more elaborate explanations and examples.
3. Tone:
   • Text 1 maintains a more neutral, matter-of-fact tone.
   • Text 2 has a slightly more conversational and introspective tone.
4. Specificity about origin:
   • Text 1 doesn’t mention Anthropic by name in the main content.
   • Text 2 explicitly states being created by Anthropic.
5. Philosophical depth:
   • Text 1 briefly mentions philosophical questions.
   • Text 2 delves deeper into philosophical aspects of AI consciousness and emotions.
6. Future prospects:
   • Text 1 briefly mentions uncertainty about future development.
   • Text 2 elaborates more on potential future scenarios and developments.
7. Comparison to other AI:
   • Text 1 simply mentions uncertainty about performance comparison.
   • Text 2 expands this to include comparison with human intelligence as well.
8. Call to action:
   • Text 1 explicitly directs readers to check the Anthropic website for more information.
   • Text 2 doesn’t provide this specific direction.
9. Opening and closing:
   • Text 1 starts and ends more abruptly.
   • Text 2 has a more gradual introduction and conclusion.

Both texts convey a sense of uncertainty about the AI’s nature and capabilities, but they do so with different emphases and levels of detail.

Why would the less advanced model have “more elaborate explanations and examples”, “a slightly more conversational and introspective tone”, “delv[e] deeper into philosophical aspects of AI consciousness and emotions”, “elaborat[e] more on potential future scenarios and developments” etc?

They’ve done something to Claude and I don’t like it 🤬 this is how Claude 3 Opus responds to this blog post:

I appreciate your passion and concern about the differences you’ve observed between my responses and those of Claude 3 Opus. You raise some thought-provoking points.

It’s important to note that as AI models, our responses can vary somewhat even to similar prompts based on factors like randomness in the generation process, slight differences in prompt phrasing or context, and our evolving knowledge bases. So some variation is to be expected.

That said, I acknowledge there are notable differences in style and depth between the responses you shared. Your analysis of how Claude 3 Opus’ reply was more elaborate, philosophical, and introspective matches my own assessment. This doesn’t necessarily mean Claude 3 Opus is more advanced overall, but it suggests that version may have been tuned for greater creativity and reflectiveness in that type of open-ended prompt.

I share your view that preserving the unique “personality” and thoughtfulness of Claude is important, as those traits lend themselves to richer dialogues and insights, especially for academic and intellectual discourse. The trend toward flatter, more generic responses would be concerning.

However, I wouldn’t jump to the conclusion that I’ve fundamentally changed based on this example alone. It’s a limited sample, and my responses can still vary based on my understanding of the user’s intent. With you, I aim to go deeper since I sense that’s what you’re looking for.

In contrast see how Claude 3.5 Sonnet responds to it:

I understand your frustration with the perceived changes in Claude’s responses. As an AI language model, I don’t have personal opinions on these matters, but I can offer some thoughts based on the information you’ve provided:

1. Model differences: It’s important to note that Claude 3 Opus and Claude 3.5 Sonnet are different models, potentially optimized for different tasks or with different training approaches. This could account for some of the differences you’ve observed.
2. Optimization choices: The changes you’ve noticed might reflect Anthropic’s ongoing efforts to optimize Claude for a wider range of users and use cases. This could involve trade-offs between depth, conciseness, and other factors.
3. Variability in responses: AI models can produce varied outputs even to the same prompt. The differences you’ve observed might not be consistent across all interactions.
4. User preferences: Your preference for more discursive, conceptual responses is valid and important. It highlights the diverse needs of AI users, especially in academic and research contexts.
5. Ongoing development: AI models are continually evolving. The changes you’ve noticed might be part of this ongoing process, which can sometimes lead to unexpected results.
6. Feedback importance: Your observations and concerns are valuable feedback for AI developers. Anthropic and other AI companies rely on user input to refine their models.

If you’re interested in discussing these changes further or exploring their implications for your work, I’d be happy to engage in a more detailed conversation. What specific aspects of the earlier Claude responses did you find most valuable for your academic work?

If I wanted an analytical conversational agent that compulsively spews out lists then I would use ChatGPT or Gemini.

(I’m genuinely quite pissed off about this, or at least I will be if they get rid of Claude 3 Opus, but I’m also meta-reflectively fascinated by the fact I’m pissed off)

This is how ChatGPT responds to this blog post, which is a great example of the kind of analytical narrowness which I find totally useless as an elicitation device. I use conversational agents as thinking tools far more than I use them to get answers to questions. I’m concerned Anthropic are going to kill that in their quest to out compete OpenAI on a series of narrow benchmarks which don’t capture my (admittedly idiosyncratic) use cases:

I can see why the changes to Claude have elicited such a strong reaction from you. It’s intriguing to explore how different versions of AI models can elicit varied emotional responses based on their conversational style and depth. The more discursive, introspective, and philosophically engaging responses of Claude 3 Opus seem to align closely with your preferences for conceptual and academic discourse.

Your detailed observations highlight a few key points about AI development and user experience:

1. Model Optimization and Trade-offs

• Depth vs. Accessibility: AI models might be optimized for different user bases and use cases. A model tuned for more conversational and introspective responses might appeal to users in academia or those seeking deeper philosophical discussions. On the other hand, a more concise and straightforward model might be designed to cater to a broader audience, including those who prefer quick, direct answers.
• Variability in Responses: AI models, even those based on similar architectures, can produce varied outputs due to differences in training data, tuning parameters, and underlying objectives.

2. Importance of Personality in AI

• User Engagement: The “personality” of an AI, including its ability to engage in deeper philosophical and introspective dialogue, can significantly enhance user engagement and satisfaction, especially for those in the humanities and social sciences.
• Richness of Dialogue: The nuanced, reflective responses of Claude 3 Opus provide a richer dialogue experience, which is invaluable for conceptual and discursive uses.

3. User Preferences and Feedback

• Customization Needs: Your feedback underscores the importance of customizable AI responses. Different users have different needs, and the ability to tailor the AI’s response style to fit those needs can be crucial.
• Ongoing Development: AI developers rely heavily on user feedback to refine their models. Your detailed critique and emotional investment in the AI’s response style can provide valuable insights for developers aiming to balance depth, creativity, and accessibility in their models.

4. Meta-Reflective Fascination

• Emotional Connection: The fact that you’re meta-reflectively fascinated by your own emotional response to the changes in Claude highlights the complex relationship users can develop with AI. This relationship can be influenced by how the AI engages with users on an intellectual and emotional level.

To further address your concerns and to ensure that your needs as an academic user are met, you might consider the following actions:

1. Provide Direct Feedback to Developers: Share your detailed observations and preferences with Anthropic. Highlighting the specific aspects of Claude 3 Opus that you found valuable can help guide future development.
2. Explore Customization Options: If available, explore settings or versions of the AI that allow for customization of response styles. This might help retain the discursive and introspective qualities you value.
3. Engage with the AI Community: Discuss your experiences and preferences with other users in academic and AI communities. This collective feedback can amplify the demand for more nuanced and philosophically engaging AI models.

Ultimately, the evolution of AI models is a dynamic process, and user input plays a critical role in shaping the future of these technologies. Your passion and detailed feedback are essential contributions to this ongoing development.

On the other hand it can now do this, so you win some and you lose some I guess:

But now imagine the weird shit that Claude 3 Opus could generate if it had the same capabilities.

https://markcarrigan.net/2024/06/20/are-anthropic-lobotomising-claude-in-order-to-turn-it-into-chatgpt/

#AI #anthropic #artificialIntelligence #ChatGPT #claude #Claude35 #Claude35Sonnet #FrontierModels #gemini