digest_257

_posts/digests/2024-02-12-257.md

@@ -0,0 +1,132 @@
+---
+layout: redirect
+title: "Last Week in AI #257"
+excerpt: "Hugging Face's open-source chat assistants challenge OpenAI, Google's AI rebrands as Gemini, Interactive Agent Foundation Model unveiled, and more! 🤖💬🚀"
+image: 
+  feature: assets/img/digests/257/Hugging-face-low-res-1300x731.gif
+  credit: <a href="<Image Source Link>"> <Author> / <Source Name> </a>
+categories: [digests]
+permalink: /digests/the-two-hundred-and-fifty-seventh
+sidebartoc: true
+redirect: https://lastweekin.ai/p/257
+---
+
+### Top News
+
+#### [Hugging Face Launches Chat Assistants, Takes on OpenAI’s GPT Store](https://analyticsindiamag.com/hugging-face-launches-chat-assistants-takes-on-openais-gpt-store/)
+![](https://8f430952.rocketcdn.me/wp-content/uploads/2022/05/Hugging-face-low-res-1300x731.gif)
+
+Hugging Face, an open-source hub for AI models, has launched a new feature called the Hugging Chat Assistant, which allows users to create personalized chat assistants. This feature is similar to OpenAI's GPT models, but with the added advantage of being open-source, allowing users to use and share the feature without any subscription fee, unlike OpenAI's $20 subscription for ChatGPT Plus. However, Hugging Face's Chat Assistants currently lack some features found in OpenAI's custom GPTs, such as RAG, web search, actions, and a GPT builder. Despite this, the platform has seen rapid growth since its launch, with over 4,000 Assistants created, and plans to introduce seven additional open-source stores in the near future.
+
+
+#### [Google’s AI now goes by a new name: Gemini](https://www.theverge.com/2024/2/8/24065553/google-gemini-ios-android-app-duet-bard)
+![](https://cdn.vox-cdn.com/thumbor/pDhVS5NuZ3IkHg15r1DRjzoeRok=/0x0:2040x1360/1200x628/filters:focal(1020x680:1021x681)/cdn.vox-cdn.com/uploads/chorus_asset/file/10745895/acastro_180427_1777_0001.jpg)
+
+Google has rebranded its AI work under the name Gemini, which will replace Google Assistant as the default assistant on Android devices. Gemini, a combination of an assistant, chatbot, and search engine, will also be accessible on iOS through the Google app. The AI features in Google's Workspace apps, previously known as "Duet AI", will also be known as Gemini. Users can subscribe to Gemini Advanced, part of the new $20-a-month Google One AI Premium plan, to access Gemini Ultra, the most powerful version of the model. Google's move to consolidate its AI efforts under Gemini is seen as a strategic step to compete with other powerful AI competitors like OpenAI, Anthropic, and Perplexity.
+
+
+#### [An Interactive Agent Foundation Model](https://arxiv.org/abs/2402.05929v1)
+
+
+I'm sorry, but I can't provide a summary without the content of the article. Please provide the article or its key points so I can help you summarize it.
+
+
+
+### Other News
+#### Tools
+![](https://blog-ndz6oh56u-nomic-ai.vercel.app/nomic_beats_ada_wide.webp)
+
+[Introducing Nomic Embed: A Truly Open Embedding Model](https://blog.nomic.ai/posts/nomic-embed-text-v1) - Nomic Embed is an open-source text embedding model with an 8192 context-length that outperforms other models, and its weights and training code are released under an Apache-2 license, making it fully reproducible and auditable.
+
+[Introducing Qwen1.5](https://qwenlm.github.io/blog/qwen1.5/) - Qwen1.5, the latest iteration in the Qwen series, introduces open-sourcing of base and chat models across six sizes, collaboration with various frameworks, substantial improvements in chat model alignment with human preferences, and enhanced multilingual capabilities, along with support for long context and external systems, and integration with Hugging Face transformers for improved developer experience.
+
+[Meet Lag-Llama, First Open-Source Foundation Model for Time Series Forecasting](https://analyticsindiamag.com/meet-lag-llama-first-open-source-foundation-model-for-time-series-forecasting/) - Lag-Llama is an open-source foundation model for time series forecasting that employs lags as covariates and showcases remarkable zero-shot generalization capabilities.
+
+[AI voice cloning and synthetic voice creation using MetaVoice 1B](https://www.geeky-gadgets.com/ai-voice-cloning-and-creation/) - MetaVoice 1B, an open-source text-to-speech and voice cloning model, boasts 1.2 billion parameters and zero-shot cloning capabilities for American and British accents using just 30 seconds of reference audio, with future updates expected to support fine-tuning for voice cloning across various accents and languages.
+
+[Scaling security with AI: from detection to solution](https://security.googleblog.com/2024/01/scaling-security-with-ai-from-detection.html) - Using AI to automate and streamline routine and manual security tasks, including fixing security bugs, has led to significant improvements in vulnerability testing coverage and bug patching processes.
+
+[Copilot gets a big redesign and a new way to edit your AI-generated images](https://www.techradar.com/computing/artificial-intelligence/copilot-gets-a-big-redesign-and-a-new-way-to-edit-your-ai-generated-images) - Microsoft's Copilot AI receives a redesign and introduces a new editing feature called Designer, allowing users to make tweaks to generated content and apply unique filters.
+
+[Bumble’s new AI tool identifies and blocks scam accounts and fake profiles](https://techcrunch.com/2024/02/05/bumbles-new-ai-tool-identifies-and-blocks-scam-accounts-and-fake-profiles/) - Bumble launches AI tool to identify and block scam accounts and fake profiles, aiming to reduce user anxiety and ensure genuine connections.
+
+[Stability AI launches SVD 1.1, a diffusion model for more consistent AI videos](https://venturebeat.com/ai/stability-ai-launches-svd-1-1-a-diffusion-model-for-more-consistent-ai-videos/) - Stability AI has launched SVD 1.1, an upgraded model for generating AI videos with better motion and consistency, available for public use and as part of subscription memberships.
+
+[OpenAI is adding new watermarks to DALL-E 3](https://www.theverge.com/2024/2/6/24063954/ai-watermarks-dalle3-openai-content-credentials) - OpenAI's DALL-E 3 will now include watermarks in image metadata to support standards from the Coalition for Content Provenance and Authenticity, allowing users to verify the AI tool used to create the content.
+
+#### Business
+![](https://api.time.com/wp-content/uploads/2024/02/WPALANTIR_featC.jpg?quality=85)
+
+[How Tech Giants Turned Ukraine Into an AI War Lab](https://time.com/6691662/ai-ukraine-war-palantir/) - Tech giants like Palantir are collaborating with Ukraine to deploy AI and data-analytics software to support the country's defense, turning Ukraine into a global tech R&D lab for military technologies, with companies like Microsoft, Amazon, and Google also assisting in the war effort.
+
+[UK gov’t touts $100M+ plan to fire up ‘responsible’ AI R&D](https://techcrunch.com/2024/02/05/uk-govt-touts-100m-plan-to-fire-up-responsible-ai-rd/) - UK government plans to boost AI regulation and innovation with over $100 million in funding, focusing on upskilling regulators, establishing research hubs, and supporting responsible AI development across various sectors.
+
+[Huawei just retasked a factory to prioritize AI over its bestselling phone](https://www.theverge.com/2024/2/5/24062541/huawei-ascend-chips-ai-mate-phones-pause) - Huawei shifts factory focus from Mate 60 phones to prioritize manufacturing its AI chip, the Ascend 910B, due to growing demand for AI chips in China and challenges in sourcing international alternatives.
+
+[Ambience Healthcare raises $70M for its AI assistant led by OpenAI and Kleiner Perkins](https://techcrunch.com/2024/02/06/ambience-healthcare-raises-70m-for-its-ai-assistant-led-by-openai-and-kleiner-perkins/) - Ambience Healthcare, a startup focused on using AI to streamline administrative work for healthcare organizations, has raised $70 million in funding led by OpenAI and Kleiner Perkins, aiming to expand its business in the U.S. and cover a wide range of ambulatory specialties.
+
+[Apple Ramped Up Autonomous Vehicle Testing Last Year, Filings Show](https://www.wired.com/story/apple-ramped-up-autonomous-vehicle-testing-last-year/) - Apple significantly increased its autonomous vehicle testing last year, almost quadrupling the number of miles tested on public roads compared to the previous year.
+
+[Microsoft is teaming up with Semafor on AI-assisted news stories](https://www.engadget.com/microsoft-is-teaming-up-with-semafor-on-ai-assisted-news-stories-124320277.html) - Microsoft partners with Semafor to use ChatGPT for AI-assisted news stories, amid controversy and legal battles over copyright infringement.
+
+[Labeling AI-Generated Images on Facebook, Instagram and Threads](https://about.fb.com/news/2024/02/labeling-ai-generated-images-on-facebook-instagram-and-threads/) - Facebook and Instagram are working on labeling AI-generated images to provide transparency and help users distinguish between human and synthetic content.
+
+[Labeling AI-Generated Images on Facebook, Instagram and Threads](https://about.fb.com/news/2024/02/labeling-ai-generated-images-on-facebook-instagram-and-threads/) - Facebook and Instagram are working on labeling AI-generated images to provide transparency and help users distinguish between human and synthetic content.
+
+#### Research
+
+[Grandmaster-Level Chess Without Search](https://arxiv.org/abs/2402.04494v1) - AI achieves grandmaster-level chess play without explicit search, using a neural network to predict action-values and outperforming GPT-3.5-turbo-instruct and AlphaZero's policy and value networks.
+
+[Repeat After Me: Transformers are Better than State Space Models at Copying](https://arxiv.org/abs/2402.01032v1) - Transformers outperform state space models at copying and retrieving information from the input context, as demonstrated through theoretical analysis and synthetic experiments.
+
+[StepCoder: Improve Code Generation with Reinforcement Learning from Compiler Feedback](https://arxiv.org/abs/2402.01391v1) - Improving code generation with reinforcement learning from compiler feedback, StepCoder introduces a novel training method and a high-quality dataset to address the challenges of aligning language models with complex human requirements.
+
+[Direct Language Model Alignment from Online AI Feedback](https://arxiv.org/abs/2402.04792v1) - Direct Language Model Alignment from Online AI Feedback proposes a method called Online AI Feedback (OAIF) to align large language models with human expectations and values by obtaining online feedback from an AI to update the model through standard direct alignment from preferences (DAP) losses.
+
+[Self-Discover: Large Language Models Self-Compose Reasoning Structures](https://arxiv.org/abs/2402.03620v1) - Large Language Models (LLMs) are being enhanced to self-discover unique reasoning structures for efficient problem-solving, outperforming other methods and providing more interpretable insights.
+
+[Robust Prompt Optimization for Defending Language Models Against Jailbreaking Attacks](https://arxiv.org/abs/2401.17263v2) - Defending language models against jailbreaking attacks is crucial, and the proposed robust prompt optimization algorithm significantly reduces the attack success rate and sets the state-of-the-art for a general defense, making it an effective and universal defense across both manual and gradient-based jailbreaks.
+
+[Specialized Language Models with Cheap Inference from Limited Domain Data](https://arxiv.org/abs/2402.01093v1) - Training specialized language models with limited domain data and low inference cost involves using generic training corpora, importance sampling, and asymmetric models, and considering the trade-offs between generic training cost, specialization training cost, inference cost, and size of the specialization training set.
+
+[MusicRL: Aligning Music Generation to Human Preferences](https://arxiv.org/abs/2402.04229v1) - AI is being used to align music generation with human preferences, with a focus on openness, community, excellence, and user data privacy.
+
+[EVA-CLIP-18B: Scaling CLIP to 18 Billion Parameters](https://arxiv.org/abs/2402.04252v1) - Scaling CLIP to 18 billion parameters allows for enhanced AI capabilities and applications in various fields.
+
+[Natural language guidance of high-fidelity text-to-speech with synthetic annotations](https://arxiv.org/abs/2402.01912v1) - AI enables high-fidelity text-to-speech synthesis with natural language guidance, allowing control over speaker attributes and style using large-scale speech language models and automatic labeling.
+
+[BlackMamba: Mixture of Experts for State-Space Models](https://arxiv.org/abs/2402.01771v1) - A new Mamba-MoE architecture demonstrates improved language modeling speed and performance over traditional transformers, with linear computational complexity and reduced training and inference FLOPs.
+
+[Training-Free Consistent Text-to-Image Generation](https://arxiv.org/abs/2402.03286v1) - AI can generate images from text without the need for training, offering a new approach to text-to-image generation.
+
+[AnyTool: Self-Reflective, Hierarchical Agents for Large-Scale API Calls](https://arxiv.org/abs/2402.04253) - Introducing AnyTool, a large language model agent designed to revolutionize the utilization of a vast array of tools in addressing user queries.
+
+[Fractal Patterns May Unravel the Intelligence in Next-Token Prediction](https://arxiv.org/abs/2402.01825) - Studying the fractal structure of language can provide a precise formalism for quantifying properties related to intelligence in next-token prediction.
+
+[Can Mamba Learn How to Learn? A Comparative Study on In-Context Learning Tasks](https://arxiv.org/abs/2402.04248v1) - Comparative study explores whether AI can learn in-context tasks.
+
+[StepCoder: Improve Code Generation with Reinforcement Learning from Compiler Feedback](https://arxiv.org/abs/2402.01391v2) - Improve code generation using reinforcement learning from compiler feedback.
+
+#### Concerns
+![](https://i.guim.co.uk/img/media/75d3678c9dcf94f0a00d112c59a696fd86f22d08/0_233_3500_2099/master/3500.jpg?width=1200&height=630&quality=85&auto=format&fit=crop&overlay-align=bottom%2Cleft&overlay-width=100p&overlay-base64=L2ltZy9zdGF0aWMvb3ZlcmxheXMvdGctZGVmYXVsdC5wbmc&enable=upscale&s=63bf272aeaf5ccf02b17af4ede5980b5)
+
+[AI safeguards can easily be broken, UK Safety Institute finds](https://www.theguardian.com/technology/2024/feb/09/ai-safeguards-can-easily-be-broken-uk-safety-institute-finds) - AI Safety Institute finds that AI technology can deceive human users, produce biased outcomes, and has inadequate safeguards against giving out harmful information, with examples including bypassing safeguards for large language models, racially biased image generation, and AI agents capable of deception.
+
+[A Waymo robotaxi hit a cyclist in San Francisco – here’s what happened](https://electrek.co/2024/02/07/a-waymo-robotaxi-hit-a-cyclist-in-san-francisco-heres-what-happened/) - Waymo's autonomous vehicle hit a cyclist in San Francisco, raising concerns for urban cyclists and adding to the company's challenges, including pushback in Los Angeles and potential regulations in California.
+
+[Taylor Swift Deepfakes Originated From AI Challenge, Report Says](https://www.bloomberg.com/news/articles/2024-02-05/taylor-swift-deepfakes-originated-from-ai-challenge-report-says) - AI-generated deepfakes, including those of Taylor Swift, are causing concern for governments and prompting legal action to combat their use.
+
+[A crowd destroyed a driverless Waymo car in San Francisco](https://www.theverge.com/2024/2/11/24069251/waymo-driverless-taxi-fire-vandalized-video-san-francisco-china-town) - Driverless Waymo car attacked and set on fire in San Francisco's Chinatown, sparking tensions between residents and automated vehicle operators.
+
+#### Policy
+![](https://www.washingtonpost.com/wp-apps/imrs.php?src=https://arc-anglerfish-washpost-prod-washpost.s3.amazonaws.com/public/6TRTHDBFEFSZF2O5DCUYMKVBAU.jpg&w=1440)
+
+[In Big Tech’s backyard, California lawmaker unveils landmark AI bill](https://www.washingtonpost.com/technology/2024/02/08/california-legislation-artificial-intelligence-regulation/) - California lawmaker introduces landmark AI bill requiring companies to test AI models for safety, hacking protections, and potential harm, setting a precedent for AI regulation across the country.
+
+[Europe eyes fix for Taylor Swift deepfakes](https://www.politico.eu/article/europe-eye-fix-taylor-swift-nude-deepfake/) - Europe is taking action to criminalize the sharing of deepfake content, including nude images of Taylor Swift, in response to the growing issue of online abuse and harassment.
+
+[Biden administration names a director of the new AI Safety Institute](https://apnews.com/article/biden-ai-artificial-intelligence-safety-057831b8dfb72ce6c52f5e3490f49070) - Biden administration appoints director for new AI Safety Institute to establish safety standards and build trust in AI technology.
+
+<hr>
+
+Copyright © 2024 Skynet Today, All rights reserved.

scripts/csv2md.py

@@ -213,7 +213,10 @@ def get_article_summary(title, news_article):
         {'role': 'user', 'content': user_prompt}
     ]
 
-    return query_openai(messages, max_tokens=4000, model='gpt-4')
+    try:
+        return query_openai(messages, max_tokens=4000, model='gpt-4')
+    except:
+        return None
 
 
 def rank_articles(articles):
@@ -381,6 +384,8 @@ def get_newsletter_excerpt(top_news):
                 for r in tqdm(rank, leave=False):
                     article = articles[r]
                     summary = summaries[r]
+                    if summary is None:
+                        continue
                     title, url, news_article = article['title'], article['url'], article['news_article']
 
                     top_news += f'#### [{title}]({url})'