[AINews] AI Discords Newsletter 11/27/2023

not like a human

                November 27, 2023

            [AINews] AI Discords Newsletter  11/27/2023

This is AI News! an MVP of a service that goes thru all AI discords/Twitters/reddits and summarizes what people are talking about, so that you can keep up without the fatigue. Signing up here opts you in to the real thing when we launch it 🔜

            Latent Space Discord Summary
Only 1 channel had activity, so no need to summarize...
Latent Space Channel Summaries
▷ Channel: ai-general-chat (6 messages):

Meetup Organization: User @ivanleomk is organizing a round of drinks for members in Singapore. The RSVP link is provided in the chat. User @eugeneyan expressed their enthusiasm for the upcoming event.
GPT captcha Issue: User @kb.v01 mentioned the recurrence of the captcha using a custom GPT, questioning if there are still scaling issues. 
Resources on GPT-4 Variants: User @jozexotic asked for resources explaining the behaviors and results of the GPT-4 variants including GPT-4-0613 and GPT-4-Turbo.
User Report: User @jozexotic reported inconsistent performance differences among the GPT-4 versions, stating the performance differences are neither 'better' nor 'worse' but dependent on the use case.
Event Announcement: User @slono shared a link to the https://codeforward.ai event asking if anyone is going. 

Links:
- Singapore meetup RSVP
- Codefoward.ai event

OpenAI Discord Summary

Discussion on AI's Consciousness and Emulation: Key topics included the possibility of attributing consciousness as a universal property, the comparison between humans and AI as different levels of consciousness emulators, the role of mimicry in perceived emotion or feeling, and the pragmatic understanding of true emotions through genuine experiences.
Detailed Debate on AI's Capabilities: Conflicting viewpoints on the categorization of models like GPT-3 as AI, and an investigation into the capabilities and limitations of AI in terms of autonomy, self-improvement, learning, skepticism, and dependence on human instruction.
Exploration of Custom-Made AI and AI Art: Discussions on how freelance designers might monetize work using AI, the commercial viability and uniqueness of AI-generated art, and contrasting opinions about the potential of AI art based on personal experience.
Inquiry about AI's Job Replacing Potential: Conversation on the potential implications of recent troubles at OpenAI on perceptions of AI's ability to replace human jobs, emphasizing the need for backup plans and alternatives due to potential instability.
Guidance Request for AI Development: User desire for learning resources expressed, though the request remained unanswered.
Delving into GPT-4 and Assistant Playground: Multiple users had discussions on enabling GPT-4 in the playground, billing concerns, and the various capabilities and limitations of the model. 
Awareness around Disallowed Uses of AI Models: Cautionary note about OpenAI's policy on disallowed usages, particularly in the context of legal purposes.
User Experiences with ChatGPT: Conversation on the application of ChatGPT in hobby-related activities, like Dungeons & Dragons game preparations and occasional server issues experienced by users while accessing the service.
Interactive Debates on AI Capabilities: Friendly debates on topics such as ChatGPT's efficiency and practicality in various contexts.
Convention on Usage Limits, Authentication and Limitations of AI: Users engaged in discussions about the usage limits of custom and standard GPTs, potential workarounds, integration of Zapier action into the GPT model, troubleshooting GPT's story continuation, user authentication concerns, and the limitations of AI models.
Dynamic Discussion on GPT-4: Chat about network issues preventing GPT-4 access, model output and token limits, troubleshooting ChatGPT issues, adoption of character traits by GPT agents, file import limits for Custom GPTs, and various introductions and creations of new GPT models.
Prompt Engineering Discourse: Points discussed included the benefits of training GPTs with Markdown, the limitations of custom GPTs in using knowledge bases effectively, crafting controversial comments, strategies for securing information with GPTs, and exploration of persistently running scripts for complex tasks.
OpenAI API Discussions: Topics such as efficacious use of Markdown with GPT, managing GPT output in JSON and challenges with the assistant API, issues around custom GPT inefficiency, guidance on improved usage of knowledge base, issues and potential solutions related to GPT state management, and an innovative example of text adventures visualized in a strategy game.

OpenAI Channel Summaries
▷ Channel: ai-discussions (407 messages🔥):

Discussion on AI's Consciousness and Emulation: Users @simple_chad, @mikeloots, and .dooz had an engaging discussion on consciousness in AI and humans. In simple_chad's view, consciousness could be attributed as a property of the universe, and humans and AI are both emulators of that consciousness but at different levels of complexity and capabilities. .dooz raised the idea that perfect mimicry can be perceived as equivalent to actual emotion or feeling, even if the 'understanding' or consciousness behind it is not the same as in humans. In response, mikeloots emphasized that truly understanding emotions requires genuine experience of them, something AI lacks at present.

Debate on AI and its Capabilities: @venkolm argued that GPT-3 isn't AI because it isn't autonomous or self-improving. He believes AI isn't capable of genuine learning or skepticism and needs continuous instruction from humans. @ted_k, .dooz, and @simple_chad disagreed, asserting that GPT-3 and comparable models are indeed AI, citing aspects like token prediction and incorporation of massive amounts of training data as key factors.

Exploration of Custom-made AIs and AI Art: @foxy_mcdick brought up how freelance designers might monetize work using the Midjourney image generator. @venkolm and @.pythagoras exchange viewpoints on the commercial viability and uniqueness of AI-generated art, with the former skeptical and the latter optimistic based on personal experience. 

Inquiry about AI's Job Replacing Potential: @mrsyntax raised a question about whether the recent troubles at OpenAI have affected attitudes towards AI's potential to replace human jobs. @eskcanta responded with the opinion that such events highlight the uncertainties and potential instability of relying solely on AI, which could lead to an increased demand for backup plans and alternatives.

Request for Guidance on AI Development: @satoru1815, expressing a desire to return to learning AI development after a 12-year hiatus, asked the group for guidance on where to start. This received no responses during the chat history provided.

▷ Channel: openai-chatter (407 messages🔥):

GPT-4 and Assistant Playground: User @ckobasti had an extensive conversation with @rjkmelb and .bytebug about enabling GPT-4 in the Assistant playground. He learned that it requires adding credit under billing, and the GPT Plus and the API are separate products and billed separately. He also learned about various capabilities of the model, including reading from PDF files and scanning images, and the possibility of programming in any language.
List of Disallowed Uses of AI Models: @eskcanta highlighted a section of OpenAI's policy around disallowed usage, specifically with regards to the usage of the AI for legal purposes, cautioning against any usage of the AI that could be perceived as offering tailored legal advice without a qualified person reviewing the information.
ChatGPT availability issue: Different users (@3dgeekshow, @alessiogr, @aar9506, @exaltist) discussed having occasional issues logging in to ChatGPT due to busy servers but it resolved after retrying.
User discussions on using the AI to generate RPG content and other hobbies: @pruo and @lugui mentioned that they used the platform to assist with the preparation for Dungeons and Dragons games. This included the generation of character backstories, creating random encounters, and general game support.
Respectful and friendly interactions between channel members: The users had light-hearted discussions and debates regarding the AI's capabilities and limitations, and interacted in a friendly manner. This included topics ranging from ChatGPT's efficiency in auto completion to its practicality in a variety of scenarios.

▷ Channel: openai-questions (226 messages🔥):

Custom GPT Development and Usage Limit: Users @d_smoov77, @stealth2077, @crerigan, @ericturner., and @vipraz engaged in a discussion on the usage limit of custom GPTs. They identified the separate usage limits for standard and custom GPTs, with the latter being slightly lower. Strategies or workarounds such as load balancing across multiple OpenAI accounts were suggested. @solbus referred to a list of billable actions that help understand the usage calculation in more detail.

Inserting Zapier Action into GPT: @danielnadeo_04361 asked for assistance in integrating a Zapier action into a GPT model. @thesocraticbeard suggested making an OpenAPI specification or adjusting instructions based on URL usage. Finally, they recommended using OpenAI's Assistants API to bring a GPT-like solution to a website, sharing the assistants overview link for reference.

GPT-Powered Story Continuation: @stealth2077 sought advice for getting GPT to continue a story. They reported that the Agent reads from somewhere in the middle of the uploaded text file rather than the end. To circumvent this, @thesocraticbeard suggested re-wording the instructions to make the agent start from the end of the text file.

GPT User Authentication Concerns: @phazei asked if deleting their OpenAI account entirely would help change the authentication method, citing the difficulty logging into Google Accounts. Several users suggested contacting OpenAI support for resolution, and @eskcanta highlighted the support@openai.com email address.

Technical Limitations of AI: @thesocraticbeard and @zwischenzugs drew attention to AI limitations, such as the inability of the AI model to directly access the internet or process tasks such as video editing or creating graphs from data. They also discussed the inability of the base GPT to modify its knowledge continually after initial training.

Other Discussions: Various other conversations touched on issues such as a deactivated account, difficulties in using DALL-E, volume issues with the OpenAI call feature, and errors encountered while using the ChatGPT API. Some users also shared their experiences and obstacles while testing or developing their custom GPT models.

▷ Channel: gpt-4-discussions (76 messages):

Network issues and restricted GPT-4 interface access: User @liming_60085 experienced network issues that restricted access to GPT-4. User @solbus redirected the user to other channels for discussion and troubleshooting.
Discussion on GPT Model Outputs and Token Limits: @truongtoan113 and @lumirix had a conversation about the number of tokens output by different GPT models. They discuss how certain models may be constrained by the CONTEXT WINDOW in outputting their responses but couldn't find documentation to back this up.
Troubleshooting ChatGPT issues: @mdehoque encountered trouble querying PDF files uploaded to ChatGPT, with the system getting stuck at "Regenerate". 'Refreshing the page' was suggested as a solution, but the problem persisted.
Instructing GPT agents to adopt character traits: @patrickgamer was improving the behavior of a GPT agent to adopt the personality traits of a Ferengi from Star Trek. After adding a bullet point emphasizing the agent to be "not like a human", a considerable change in the response was noticed.
File import limits for Custom GPT: @solbus informed @snowzer that up to ten 512MB files can be imported into Custom GPT, providing a link to more information on the topic.
Introductions and New GPT Creations: @tunter. and @mikepaixao introduced themselves and shared links to their created GPTs such as the "Lex Fridman Podcast" and "Great Sage".
Users @marta.m.p, @kolehti, and @satanhashtag discussed the current wait time for access to ChatGPT 4. There was no definitive answer or ETA available.
@dotails brought up a suggestion for adding the ability to switch back to GPT-4 from GPT 3.5 once the wait time is over. This feature is currently not available.

▷ Channel: prompt-engineering (25 messages):

Training GPTs with Markdown and Plain Text: In a discussion initiated by @jungle_jo, users highlighted that GPT has been trained on Markdown, and that using Markdown to guide it can be beneficial. @eskcanta also added that language fluency between user and AI contributes to effective communication. @fran9000 and @eskcanta further discussed the use of plain text and line breaks in knowledge base files.
Limitations of GPTs: @johnz999 voiced concerns about inefficiencies in custom GPTs, specifically regarding their infrequent use of knowledge bases and incoherent responses.
Controversial Comments: @bambooshoots critically addressed unknown user's lack of understanding about domain knowledge, prompt engineering, and LLMs.
Securing Information with GPTs: @eskcanta explained the complexities involved in training an AI to both acknowledge and not share certain secrets. 
Persistent Python Script Possibilities: A conversation led by @fran9000 and @eskcanta explored the idea of a persistently running Python script that stores state data, especially for more complex or conditional tasks.

▷ Channel: api-discussions (25 messages):

GPT and Markdown: @jungle_jo queries about the effective use of Markdown with GPT. @eskcanta suggests using any language both the user and AI understand well.

GPT output in JSON format: @boomboom68 asks for prompts to restrict GPT-4 to produce output in JSON because the assistants API lacks a JSON mode.

Custom GPT inefficiency: @johnz999 comments on custom GPTs inefficiency, mentioning their issues with referencing the knowledge base correctly. Response by @bambooshoots criticizes these claims due to user misunderstanding and lack of expertise.

Knowledge Base Guidance: @eskcanta shares insights on checking the knowledge base, providing clear instructions and avoiding conflicts. Also mentions their ongoing work on an improved SecretKeeperGPT aimed at resisting all possible trickery.

GPT state management: @fran9000 and @eskcanta discuss the challenge of managing GPT states and the value of getting/saving state data on a server. @fran9000 also shares the details of their GPT based on the classic TV show "The Prisoner". 

Visualization of a text adventure: @stealth2077 provides an example of an in-text scenario for a strategy game, aiming to get GPT to visualize stats and not need to be reminded to generate stats and images.

LangChain AI Discord Summary

Update on langchain-core by @bagatur, noting the release of version 0.0.341 and its dependency on langchain-core 0.0.6, with advice against direct usage of langchain-core until the 0.1 release. Source
Discussion on the best strategy for using Pinecone as a vector database backend, considering either creating a new pod for each user or using two large pods and namespaces.
Reported issue with Inference API timeout on HuggingFaceHub when trying to use a non-listed model for a RAG application.
Inconsistencies with the SERP API used with agents, specifically yielding incorrect responses for certain basic queries.
Issue with a Data Manipulation Agent terminating at a Thought instead of reaching the Final answer during JSON return operations.
Consultation on building a chatbot with context-relevant follow-up capabilities, with recommendation to tailor the prompt accordingly.
Inquiry on the correct procedure to scrape selective elements of a website using UnstructuredURLLoader() in element mode.
Query on how to retrieve the response status code and specific HTTP headers from a non-200 response from the OpenAI API.
Experienced issue with RunnablePassthrough in LangServe, which causes the disappearance of original fields, with a workaround provided.
Inquiry and interest expressed towards adding Django support to Langserve for larger, scalable applications with ORM capability.
Enablement of callback event sending from a LangServe server, with a linked parameter, and notice of limited testing and support only for invoke and batch.
Solutions for unexpected behave by type inference, with recommendation to use the with_types function.
Instructions for setting up a Chat Widget in LangServe, with notice of potential future changes to output key behavior. Chat Widget Setup
Interest in keeping systems self-hosted and using MongoDB with Retriever Docstore, with successful document storage reported.
Inquiry on how to set up a Chainlit frontend using the Research Assistant template.
Request for assistance in specifying the {lang} parameter in a Conversational Retrieval QA Chain and issues with language output, even after manually setting the language.
Sharing of a Docker setup for the LangServe Research Assistant Template, accessible from a GitHub link with future plans outlined. Docker Setup
Sharing of a thread on creating a RAG Chain with Sources using the LangChain expression language. RAG Chain with Sources

LangChain AI Channel Summaries
▷ Channel: announcements (1 messages):

Update on langchain-core: @bagatur announced the release of the first version of langchain (0.0.341) that utilizes langchain-core (0.0.6). They mentioned that this update shouldn't affect the usage of langchain and advised not to switch application code to using langchain-core directly until the release of langchain-core 0.1, which is scheduled for 12/8. Users have been urged to report any issues they encounter with the new release.

▷ Channel: general (17 messages):

Use of Pinecone as Vector Database Backend: @sullynaj asked about the best approach to using Pinecone as the vector database backend in their platform. Specifically, they are considering whether to create a new pod for each new user or to use two large pods and namespaces for all users.
Inference API timeout with HuggingFaceHub: @kamlesh experienced an Inference API timeout when trying to use a model not available in langchain's HuggingFaceHub list for a RAG application.
SERP API Usage with Agents: @sid.pocketmail was having inconsistencies with the SERP API when used with agents. They reported receiving incorrect responses for certain simple questions.
Data Manipulation Agent Stopping at a Thought: @_lawgic is leveraging an agent for data manipulation tasks intended to return JSON at each step. They have encountered an issue where the agent sometimes ends at a Thought instead of reaching the Final answer.
Building a Chatbot with Follow-Ups: @._.nobody._._ sought advice on developing a chatbot capable of asking contextually relevant follow-up questions. @seththunder suggested tailoring the prompt to include an instruction for asking follow-up questions.
Web Scraping Query: @diegoru98 asked for guidance on scraping selective elements from a website using the UnstructuredURLLoader() in element mode. 
Retrieve Response Status and HTTP headers: @wyang14 asked how to fetch the response status code and specific HTTP headers from a non-200 response from the OpenAI API.

▷ Channel: langserve (7 messages):

Issues with RunnablePassthrough in LangServe: User @abalog_92335 reported an issue where running a RunnablePassthrough with LangServe causes original fields to disappear. A solution was found by placing a RunnablePassthrough() before and after the main chain. The user expressed a desire to understand why this solution works and how to use it in a more "proper" way.
Interest in Django Support: User @veryboldbagel inquired about the community's interest level in adding Django support, and @aliikhatami94 expressed a lot of interest to support large, scalable applications with an ORM capability.
Enabling Callback Event from the Server: @veryboldbagel highlighted the ability to enable callback event sending from the server using this parameter and admitted its feature is not well tested. It was also mentioned that this is only supported for invoke and batch.
Solving Type Inference Issue: To address a situation where type inference acted unexpectedly, @veryboldbagel suggested using the with_types function to specify the expected inputs, as demonstrated in the mentioned code snippet.
Chat Widget Setup: @veryboldbagel provided a link to the code for setting up a chat widget in LangServe but cautioned users that the behavior of the output key will likely change in the future, resulting in a breaking change. The link to set up a chat widget is here.

▷ Channel: langchain-templates (4 messages):

Self-hosting and Using MongoDB with Retriever Docstore: User @alex_35579 expressed a desire to keep things self-hosted and expressed interest in directly referring to MongoDB in the Retriever Docstore for convenience. They confirmed that they had MongoDB working and were able to store documents within it.

Setting up a Chainlit Frontend with Research Assistant Template: User @0ymika inquired about how to set up a Chainlit frontend utilizing the Research Assistant template.

Code Help for Specifying Language Parameter in Conversational Retrieval QA Chain: User @menny9762 requested assistance with their Javascript code for a Conversational Retrieval QA Chain. They were specifically struggling with specifying the {lang} parameters and also mentioned that even when they manually set the language, the response often came in English. The said code, including the QA_PROMPT constant and creation of the Retrieval QA Chain, was provided within the post.

▷ Channel: share-your-work (1 messages):

Docker Setup for Langserve Research Assistant Template: @joshuasundance has shared a simple Docker setup to experiment with the Research Assistant Template using Langserve. The Docker setup can be accessed from this GitHub link. Future plans include pushing the setup to DockerHub and adding an OpenAI key to the Kubernetes yaml file.

▷ Channel: tutorials (1 messages):

RAG Chain with Sources: User @wassup6526 shared a thread about how to create a RAG Chain with Sources using the LangChain expression language. The thread can be accessed through the provided link https://x.com/waseemhnyc/status/1728889388954951717?s=20.

Nous Research AI Discord Summary

In-depth discussions on AI model performance and training strategies occurred across the guild:

@euclaise discusses balancing subjective performance gains with potential benchmark score losses during model training. 
@danielpgonzalez proposes a method for improving beam search run in parallel and evaluates using a Large Language Model. 
@yorth_night suggests fine-tuning a base Arabic model on the Moroccan dialect for models designed for low resource languages. 
@giftedgummybee advises limited epochs to avoid overfitting on data-constrained datasets.
@teknium recommends tried and tested model parameters for running latest OpenHermes models in LM Studio. 

Multiple new AI Models such as Goliath, Hermes, and Koishi were highlighted by @ludis___ and @nonameusr, detailing their individual capacities and potential combinations.

@nonameusr initiated an insightful dialogue on enhancing AI teachability and experiential learning, highlighting strategies like iterative learning, self-initiated thought, and improved information retrieval.

Usage of Reward Models in AI training was examined, with @tekniumendorsing their enhanced scoring capabilities. @casper_ai revealed successful application of Starling-RM-7B-alpha reward model to refine Mistral 7B.

Model comparisons noted by @orabazes and @variav3030, favoring OpenHermes over variants of Capybara.

Practical issues raised included @kdawgdfw seeking advice on llama.cpp, @nonameusr inquiring about accessing Huggingface's training datasets, and @latentfog querying about multiple evaluation datasets in Huggingface trainers.

Interesting issues and ideas outside the technical domain, such as dream experiences of technical work, the desire for brand swag, and advice on avoiding travel scams encountered while studying low resource languages.

A plethora of topical links was shared across multiple topics, like AI research papers, GitHub repositories, YouTube videos, and tweets. These inclusive of specific links like Mergekit Tool, cDPO Technique Paper, and Starling-RM-7B-alpha.

Nous Research AI Channel Summaries
▷ Channel: ctx-length-research (2 messages):

Expressivity of the Attention Mechanism: @euclaise mentioned that reducing the expressivity of the attention mechanism might not have any negative impact.
RNNs Before Transformer: @euclaise suggested a solution of adding RNNs before the transformer and shared a link to a related research paper titled "Recurrent Transformers for Speech Recognition".
Link: Recurrent Transformers for Speech Recognition Paper

▷ Channel: off-topic (18 messages):

Formatting of Hermes: @teknium pointed out an issue with the formatting in Hermes.
Scraping Pipeline Issue: @tsunemoto mentioned that something was odd with his scraping pipeline and that they are currently working on it.
Nous Research Stickers: Both @imchrismayfield made a request for Nous Research stickers for their laptops, which @teknium promised to bring to the Ollama event.
Dream and Nightmare Discussions: There were discussions about dreams related to technical work. @yorth_night shared that they dream about full arXiv papers and @somewheresy and @yorth_night mentioned nightmares about Node dependency and CORS issues.
Links shared:
YouTube video shared by @fullstack6209
YouTube video shared by @pradeep1148

▷ Channel: interesting-links (9 messages):

User @fullstack6209 shared multiple resources:

A guide on dealing with limited GPU resources: GPU for the Poor
An OpenReview research paper: OpenReview Paper
A YouTube video: YouTube Video
A GitHub repository named IJEPA by Facebook Research: IJEPA

User @_automagic provided two links:

An unspecified link: Link
A tweet found interesting: Tweet

@firefox8975 shared a Twitter link: Tweet

@metaldragon01 shared a tweet from user banghuaz: Tweet

▷ Channel: general (211 messages🔥):

Model Performance and Training Discussions: @euclaise discussed the training of AI models, mentioning that training on one prompt format might reduce benchmark scores but improve subjective performance. @danielpgonzalez proposed a method improving beam search by running multiple beams in parallel and using an LLM (Large Language Model) to evaluate the quality of each beam. 
Goliath, Hermes, and Koishi Models: @ludis___ introduced Goliath, a 120b model produced by merging two Llama2 70b models. @nonameusr described initial tests done on Goliath and recommended fixing censorship issues. @ludis___ confirmed plans to fix this issue and also revealed the name of another AI model in development, called Koishi. The idea of merging the OpenHermes 2.5 with Nous Capybara 34b and Koishi to create a 161b model was suggested.
Discussion on AI Teachability and Experience: Various users, namely @nonameusr, used the chat to discuss about making AI models more teachable and aware. The discussion covered ideas such as reiterative learning, self-initiated thought, linking previous lessons to create a more cohesive educational experience for the user, and enhancing creativity and information retrieval simultaneously.
Reward Model Discussion: A conversation took place regarding the use and efficacy of Reward Models in AI training. @teknium suggested that reward models were capable of better scoring and reasoning. @thepok suggested trying out APA, PPO, and P3O techniques and noted that online training yields better results than offline training, despite being more challenging. @casper_ai mentioned the release of the Starling-RM-7B-alpha reward model, and highlighted their use of this reward model to fine-tune Mistral 7B.
Links of Interest: Several links and resources were shared:
Mergekit Tool
Reddit Link for LLM Comparison Test
cDPO Technique Paper
Tweet on MT-Bench Score

▷ Channel: ask-about-llms (63 messages):

Performances of Different Models: Various users including @orabazes and @variav3030 discussed the effectiveness of different AI models. It was stated that certain versions of OpenHermes appear to perform better overall compared to Capybara-Yi and Capybara-Tess-Yi-200k-dare-ties. @_automagic noted that LoRa performed slower when sufficient VRAM is available, as confirmed in a shared link. 
Downloading Data from Huggingface: @nonameusr inquired about downloading training data for models on Huggingface. @yorth_night explained that this can be done by accessing the datasets a model was trained on, provided they have been made public by the model creator. 
Multiple Evaluation Datasets in Huggingface Trainers: @latentfog sought advice on how to use multiple evaluation datasets in Huggingface trainers, noting that it seems accelerate is the only available option currently.
Efficient Training for Low Resource Languages: @yorth_night and @russselm had an extended discussion about developing a model to understand the Moroccan dialect. They decided on starting with a base model trained on Arabic and then finely tuning this on the Moroccan dialect. Several epochs were suggested for this training, with @giftedgummybee advising a final set of 4 epochs for data-constrained datasets to avoid overfitting. The idea of synthetic data generation was also discussed, alongside tips for avoiding scams while travelling in Morocco.
Running LLMs with Llama.cpp: @kdawgdfw requested advice on working with llama.cpp, in particular about the ideal command line arguments for running the latest OpenHermes. @teknium recommended default arguments used in LM Studio.

Alignment Lab AI Discord Summary

In the ai-and-ml-discussion channel, @hoshi_hiyouga brought up the use of a re-rank technique. In response, @spirit_from_germany requested more context or information on the topic. 
In the oo channel, user @cryptossssun shared a research paper: https://arxiv.org/pdf/2311.12023.pdf.
There was no noteworthy discussion in the general-chat channel to summarize.

Alignment Lab AI Channel Summaries
▷ Channel: ai-and-ml-discussion (2 messages):

@hoshi_hiyouga mentioned the use of a re-rank technique in an unspecified context.
@spirit_from_germany requested a link for more context or additional information on the topic.

▷ Channel: general-chat (1 messages):
The channel has no messages for the summarizer AI to summarize.
▷ Channel: oo (1 messages):

A user @cryptossssun shared a link to a research paper: https://arxiv.org/pdf/2311.12023.pdf.

Skunkworks AI Discord Summary
Only 1 channel had activity, so no need to summarize...
Skunkworks AI Channel Summaries
▷ Channel: off-topic (1 messages):

A user, @pradeep1148, shared a YouTube link: https://youtu.be/Fff0Cl9_bMc. The topic or content of the video was not discussed.

LLM Perf Enthusiasts AI Discord Summary

A speculative but potentially unhinged and untrue document on Q* was shared by @res6969 in the Google Docs Link.

Discussion related to finding a faster alternative to SerpAPI initiated by @23goat, with @potrock suggesting Metaphor as a possible solution, though noted that queries would have to be redone. A link was provided by @firefox8975.

@rabiat shared a job opportunity for an AI Engineer at SynthFlow.ai, a company working on AI accessibility and building AI voice agents for B2B clients. The AI Engineer job post is available for more details.

A conversation about real-life meetups in New York City was initiated by @robotums and @.psychickoala, with an interest shown by @minimario. @ivanleomk also suggested a meetup in Singapore and shared an RSVP link for the event.

Intricate discussions about GPTs and prompt engineering, with @jmtqjmt voicing concerns over their approximately 15 GPTs' sensitivity to prompt engineering. A YouTube video for additional insights and a cookbook tutorial suggested by @daymanfan for understanding the mentioned video.

LLM Perf Enthusiasts AI Channel Summaries
▷ Channel: opensource (1 messages):
Apologies, but the message from @joschkabraun is incomplete and no specific discussion or topic has been provided to be summarized. Without complete information, a summary cannot be provided.
▷ Channel: offtopic (5 messages):

Q Speculation Document: User @res6969 shared a Google document related to speculation on Q. The user noted that the speculation is potentially unhinged and untrue but found it interesting.

▷ Channel: speed (3 messages):

Searching for faster alternatives to SerpApi: User @23goat asked for a faster alternative to SerpAPI as it currently takes ~2-3s to retrieve the top 5 links for a query.
Suggestion for Metaphor as an alternative: @potrock suggested using Metaphor instead of Google serp but noted that queries would have to be redone.
Link shared by firefox8975: A link was shared by @firefox8975.

▷ Channel: jobs (1 messages):

Job Opening at Synthflow.ai: User @rabiat shared that Synthflow.ai is hiring an AI Engineer to join their team to lead product development, research, and technical architecture for their new AI-powered platform. The company focuses on making AI accessible to everyone, and their platform helps B2B clients build AI Voice agents.
Link: AI Engineer job post

▷ Channel: irl (5 messages):

New York City Meetup Discussion: Users @robotums and @.psychickoala initiated a conversation about hosting a meetup in New York City. User @minimario expressed interest but mentioned a planned trip to Singapore in early December.
Singapore Meetup Planning: User @ivanleomk proposed organizing a meetup for Singapore users and shared an RSVP link for the event. You can access it here.

▷ Channel: openai (3 messages):

Use of GPTs and prompt engineering: In a conversation, @jmtqjmt mentioned the high sensitivity of their approximately 15 GPTs to prompt engineering, and queried the approach to actions though they haven't explored them in detail.
Video Resources: 
@jmtqjmt is also following a YouTube video for additional insights.
@daymanfan pointed out the relevance of a cookbook tutorial to the video mentioned by @jmtqjmt.

MLOps @Chipro Discord Summary
Only 1 channel had activity, so no need to summarize...
MLOps @Chipro Channel Summaries
▷ Channel: events (2 messages):

Webinar: AI & IIoT Unlocking Industry 4.0: User @kirstenl announced a webinar happening on November 29, 2023, at 12PM ET. Speakers Seth Clark and Brad Munday from Modzy will discuss edge AI in digital transformation, the role of MQTT, reference architectures and more. The webinar is aimed at ML / AI Engineers, IoT engineers, industrial engineers, application engineers. Registration Link
Webinar: Breaking Down the Economics of LLMs in Production: User @idan_benram_62348 posted about a webinar set for December 5th, at 7:00 PM GMT. The webinar will focus on the economics of LLM deployments (cloud vs. on-premise), LLM cost structures, optimizing LLM usage for balance between speed and cost-efficiency, and measuring the unit economics of your LLM application. The webinar is ideal for business leaders and IT professionals seeking scalable, cost-effective AI solutions. Registration Link

Ontocord (MDEL discord) Discord Summary
Only 1 channel had activity, so no need to summarize...
Ontocord (MDEL discord) Channel Summaries
▷ Channel: general (1 messages):

Discussion about Infrastructure Channel: User @rvencu suggested creating an infrastructure channel to discuss related needs.

AI Engineer Foundation Discord Summary
Only 1 channel had activity, so no need to summarize...
AI Engineer Foundation Channel Summaries
▷ Channel: general (1 messages):

Tonic_1's Suggestion for Use of Allen AI Tulu Model: @tonic_1 suggests members create a free app using the Agent Factory / Agent Builder app with the Allen AI Tulu model. It hasn't been made fully compatible with agent protocol yet, but should still serve the purpose. He shared a link to the model for users interested in this project.

The Perplexity AI Discord has no new messages. If this guild has been quiet for too long, let us know and we will remove it.

The YAIG (a16z Infra) Discord has no new messages. If this guild has been quiet for too long, let us know and we will remove it.

Don't miss what's next. Subscribe to AI News (MOVED TO news.smol.ai!):