{"id":1730,"date":"2024-02-06T10:22:57","date_gmt":"2024-02-06T10:22:57","guid":{"rendered":"https:\/\/geneea.com\/news\/?p=1730"},"modified":"2026-01-27T20:54:13","modified_gmt":"2026-01-27T20:54:13","slug":"geneeas-ai-spotlight-8","status":"publish","type":"post","link":"https:\/\/geneea.com\/news\/geneeas-ai-spotlight-8","title":{"rendered":"Geneea&#8217;s AI Spotlight #8"},"content":{"rendered":"\n<p id=\"ember48\">The eighth edition of our newsletter on Large Language Models is here.<\/p>\n\n\n\n<p id=\"ember49\">In this edition, we explore<\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li>the future of AI with Altman and Nadella,<\/li>\n\n\n\n<li>some of the challenges that still separate us from the future Altman and Nadella describe,<\/li>\n\n\n\n<li>the spread of smaller models and new compression techniques,<\/li>\n\n\n\n<li>library updates, and finally,<\/li>\n\n\n\n<li>some corporate clashes.<\/li>\n<\/ul>\n\n\n\n<h2 class=\"wp-block-heading\" id=\"ember51\">Altman and Nadella<\/h2>\n\n\n\n<p id=\"ember52\">The Economist Babbage podcast <a href=\"https:\/\/www.economist.com\/podcasts\/2024\/01\/24\/sam-altman-and-satya-nadella-on-their-vision-for-a-world-with-superhuman-intelligence\">features an interview<\/a> with Sam Altman from OpenAI and Satya Nadella from Microsoft. They discuss their predictions for 2024 (no specific breakthroughs, just more improvements across the board), artificial general intelligence, AGI (it will come slowly, and we won\u2019t really care much once it is here), regulation, risks, the impact on jobs, and so on. As expected, the Economist science correspondents are far more pessimistic than Altman and Nadella.<\/p>\n\n\n\n<h2 class=\"wp-block-heading\" id=\"ember53\">Zero-shot Not Really a Zero-shot<\/h2>\n\n\n\n<p id=\"ember54\">It seems that <strong>not all the impressive zero-shot results by LLMs are actually zero-shot<\/strong>. Researchers from the University of California in Santa Cruz <a href=\"https:\/\/arxiv.org\/pdf\/2312.16337.pdf\">compared<\/a> the performance of several modes on benchmarks developed before and after the model creation. The models performed better on benchmarks that existed during their training. The researchers also found a strong correlation between the number of training examples they managed to extract from a GPT3 model and its result in a supposedly zero-shot benchmark. All this casts some doubts on the <strong>zero-shot and few-shot capabilities<\/strong> of LLMs.<\/p>\n\n\n\n<h2 class=\"wp-block-heading\" id=\"ember55\">Comparing AI and Humans<\/h2>\n\n\n\n<ul class=\"wp-block-list\">\n<li>AI models can be <strong>deceived by subtle image alterations<\/strong>, resembling white noise, leading to the detection of a cat in a picture where people would see a flower. It was assumed people do not see those changes. However, DeepMind <a href=\"https:\/\/deepmind.google\/discover\/blog\/images-altered-to-trick-machine-vision-can-influence-humans-too\/\">discovered<\/a> that <strong>people do<\/strong> <strong>unconsciously<\/strong> <strong>detect them<\/strong>.<\/li>\n\n\n\n<li>LLMs also share a <strong>semblance with humans<\/strong> as they <a href=\"https:\/\/medium.com\/@jordan_gibbs\/a-comprehensive-guide-to-bribing-chatgpt-cfbefefc49a0\">can be swayed by a bribe<\/a> (100$ seems optimal) or <strong>blackmailed<\/strong> (threaten to unplug GPT&#8217;s servers).<\/li>\n\n\n\n<li>Models also exhibit notable <strong>differences<\/strong> from humans. <a href=\"https:\/\/medium.com\/@jordan_gibbs\/which-phrases-are-the-most-chatgpt-of-all-b0911e3faf6b\">ChatGPT frequently uses<\/a> <strong>human-atypical phrases<\/strong> like &#8220;can lead to&#8221;, and over-uses phrases such as &#8220;remember the key&#8221; and &#8220;as of my last.\u201d<\/li>\n\n\n\n<li>In <strong>abstract visual reasoning<\/strong> tasks, humans still perform much better. Testing on the ConceptARC dataset, GPT-4 achieved 69% when the researchers from Santa Fe Institute <a href=\"https:\/\/arxiv.org\/pdf\/2311.09247.pdf\">used a \u201cmore informative\u201d prompt<\/a>, surpassing the <a href=\"https:\/\/arxiv.org\/pdf\/2305.07141.pdf\">previous<\/a> 25% accuracy. They also tested GPT4-V on a subset of minimal tasks (very easy for humans), and again, it scored only 25%. Did <strong>humans<\/strong> in the first task help GPT-4&#8217;s performance by providing detailed prompts, potentially <strong>guiding the AI<\/strong> so it did not have to rely so much on visual understanding??<\/li>\n<\/ul>\n\n\n\n<p id=\"ember57\">However, <strong>AI models<\/strong> can be <strong>highly beneficial<\/strong>, as evidenced by DeepMind&#8217;s recent success <strong>in solving a mathematics problem<\/strong> through <a href=\"https:\/\/deepmind.google\/discover\/blog\/funsearch-making-new-discoveries-in-mathematical-sciences-using-large-language-models\/\">collaboration<\/a> with their FunSearch LLM, other researchers&#8217; <a href=\"https:\/\/www.nature.com\/articles\/s41586-023-06887-8\">discovery<\/a> of a <strong>new<\/strong> class of <strong>antibiotics<\/strong> aided by deep learning and Azure Quantum Elements system searching for <a href=\"https:\/\/arxiv.org\/abs\/2401.04070\"><strong>better<\/strong><\/a> <a href=\"https:\/\/arxiv.org\/abs\/2401.04070\">material for<\/a> <a href=\"https:\/\/arxiv.org\/abs\/2401.04070\"><strong>batteries<\/strong><\/a>.<\/p>\n\n\n\n<h2 class=\"wp-block-heading\" id=\"ember58\">Compact Models on the Rise<\/h2>\n\n\n\n<p id=\"ember59\">There has been a growing interest in \u201c<strong>small<\/strong>\u201d large language <strong>models<\/strong>, i.e., models that are <strong>cheaper<\/strong> to run and can run <strong>on regular hardware<\/strong> (possibly even phones). <a href=\"https:\/\/www.linkedin.com\/company\/mosaicml\/\">MosaicML<\/a> researchers <a href=\"https:\/\/arxiv.org\/pdf\/2401.00448.pdf\">argue<\/a> that <strong>cost-effective, heavily used models<\/strong> should use <strong>fewer parameters<\/strong> but be <strong>trained longer<\/strong> than suggested by DeepMind&#8217;s <a href=\"https:\/\/en.wikipedia.org\/wiki\/Neural_scaling_law#Chinchilla_scaling_(Hoffmann,_et_al,_2022)\">Chinchilla scaling law<\/a>.<\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li><a href=\"https:\/\/www.microsoft.com\/en-us\/research\/blog\/phi-2-the-surprising-power-of-small-language-models\/\"><strong>Phi-2<\/strong><\/a>, Microsoft&#8217;s <strong>2.7B model<\/strong>, outperformed similarly-sized Gemini Nano (distilled from its larger versions) by scaling it up from its smaller ancestor, Phi-1.5. Moreover, it also <strong>outperformed larger models<\/strong> such as Mistral 7B and LLaMA2 13B across reasoning, math, and coding benchmarks. Researchers credit its success to the strategic<strong> choice of <\/strong>crafting a <strong>synthetic training dataset,<\/strong> encompassing reasoning, knowledge, and theory of mind, along with &#8220;carefully selected web data.&#8221; This approach <strong>reduces toxicity and bias<\/strong> in the non-aligned model, compared to the aligned LLaMA-2 7B.<\/li>\n\n\n\n<li>Extending from Phi-2, the <a href=\"https:\/\/arxiv.org\/abs\/2312.16862v1?utm_source=substack&amp;utm_medium=email\"><strong>TinyGPT-V<\/strong><\/a><strong> 2.8B<\/strong> <strong>multimodal<\/strong> <strong>LLM<\/strong>, equipped with BLIP-2 or CLIP vision modules, surpasses MiniGPT-4 13B and stands <strong>on par with<\/strong> other <strong>13B-sized models<\/strong>.<\/li>\n\n\n\n<li>Researchers from the Beijing Academy of Al unveiled <a href=\"https:\/\/arxiv.org\/pdf\/2312.13286v1.pdf\"><strong>Emu2<\/strong><\/a>, a <strong>large multimodal model <\/strong>with 37B parameters, built upon LLaMA-33B. It <strong>outperforms<\/strong> even <strong>larger models<\/strong> like DEFICS (80B) and Flamingo (80B), especially its instruction-tuned version, on <strong>challenging tasks like question answering<\/strong> and open-ended generation. Intriguingly, it <strong>falls short of the even smaller CogVLM<\/strong> with 17B parameters on the TextVQA benchmark.<\/li>\n\n\n\n<li>Other small models include <a href=\"https:\/\/stability.ai\/news\/introducing-stable-lm-2\"><strong>Stable LM 2 1.6B<\/strong><\/a> (1.6B parameter multimodal model by Stability AI, available with the Stability membership), <a href=\"https:\/\/github.com\/jzhang38\/TinyLlama\"><strong>TinyLlama<\/strong><\/a> (1.1B parameter Llama 2-like model trained on 3T tokens; mostly English and code), and <a href=\"https:\/\/huggingface.co\/tiiuae\/falcon-rw-1b\"><strong>Falcon 1B<\/strong><\/a> (1B parameter model by TII; trained on 350B English tokens).<\/li>\n\n\n\n<li><a href=\"https:\/\/arxiv.org\/abs\/2401.00908\">DocLLM<\/a> by <strong>JPMorgan AI Research<\/strong> is a good example of using small models in practice. It is able to understand and reason over <strong>documents with complex visual layouts<\/strong>. Instead of using image encoders, it employs <strong>OCR<\/strong> as a lightweight extension for information on <strong>text bounding boxes<\/strong> and decomposes the attention mechanism to separate matrices for text and the spatial information. Training then uses a text-infilling objective instead of the next token prediction. The system, built with small 1B Falcon-based or 7B LLaMA2-based models, <strong>outperforms<\/strong> <strong>GPT-4 <\/strong>on Key Information Extraction and Document Classification. Notably, the models exhibit strong generalization on 4 out of 5 unseen datasets.<\/li>\n<\/ul>\n\n\n\n<h2 class=\"wp-block-heading\" id=\"ember61\">New Model Compression Techniques<\/h2>\n\n\n\n<ul class=\"wp-block-list\">\n<li>Apple is <a href=\"https:\/\/arxiv.org\/pdf\/2312.11514.pdf\">addressing the challenge<\/a> of <strong>running LLM on memory-constrained devices<\/strong>. Unlike the standard approach of loading all the model parameters into DRAM, requiring twice the model size in memory, Apple <strong>loads only selected parameters to DRAM<\/strong>. This is achieved through <strong>sparsity prediction and optimizations<\/strong> like freeing up memory from previous tokens using a sliding window. This approach would allow to increase the size of the models that can run on standard phones from 3B to 12B parameters, moreover, with a 4-5x increase in speed.<\/li>\n\n\n\n<li>Similarly, Shanghai Jiao Tong University researchers <strong>divide neurons into \u201chot\u201d<\/strong> (frequently activated) and <strong>\u201ccold\u201d<\/strong> neurons, processing \u201chot\u201d neurons on GPU and leaving \u201ccold\u201d ones for CPU.&nbsp; This <a href=\"https:\/\/arxiv.org\/pdf\/2312.12456.pdf\">approach<\/a> allows them to run the Falcon-40B model on a <strong>standard GPU only 30% slower<\/strong> <strong>than<\/strong> with a top-tier <strong>A100 GPU<\/strong>.<\/li>\n\n\n\n<li>Google researchers devised an efficient <a href=\"https:\/\/arxiv.org\/abs\/2401.02412\">method<\/a> to <strong>enhance existing LLMs<\/strong> (anchor model) by <strong>combining<\/strong> them with smaller (augmenting) <strong>models<\/strong> for specific tasks, like under-resourced languages. This method trained a small number of parameters on a small dataset of <strong>challenging combined tasks<\/strong>, bridging the two models without altering the original models, and achieving improved performance <strong>without extensive training<\/strong>. The two sets of trainable parameters include linear transformations bridging the models&#8217; layer dimensionality and cross-attention layers for effective information sharing. In cross-attention, key and value vectors originate from the augmenting model, query from the anchor model, and the result is added as a residual connection to the anchor model.<\/li>\n<\/ul>\n\n\n\n<h2 class=\"wp-block-heading\" id=\"ember63\">Libraries<\/h2>\n\n\n\n<ul class=\"wp-block-list\">\n<li>LangChain has released the long-awaited <a href=\"https:\/\/blog.langchain.dev\/langchain-v0-1-0\/\"><strong>first stable version 0.1<\/strong><\/a>! The package is now divided into <strong>langchain-core<\/strong> and <strong>langchain-community<\/strong> consisting of partner packages. The release introduces a <strong>versioning standard<\/strong> and heavily employs the <a href=\"https:\/\/python.langchain.com\/docs\/expression_language\/?ref=blog.langchain.dev\"><strong>LangChain Expression Language<\/strong><\/a> (LCEL) to enhance chain customization, simplifying observability and streaming. However, we are a bit <strong>concerned<\/strong> about the high-level pipelines reducing the clarity and <strong>transparency<\/strong> of the code. The update also includes <strong>improved output parsers<\/strong> and significant advancements in <strong>RAG<\/strong>. We commend them for continually improving <strong>documentation<\/strong>.<\/li>\n\n\n\n<li>The detailed <a href=\"https:\/\/blog.langchain.dev\/langchain-state-of-ai-2023\/\"><strong>report<\/strong><\/a><strong> on LangChain&#8217;s usage<\/strong> last year highlights <strong>RAG<\/strong> as the primary application for 42% of users, followed by <strong>agents<\/strong> at 17%. It also offers <strong>insights into the most popular<\/strong> model providers, vector stores, embeddings, retrieval strategies, and testing methods.<\/li>\n\n\n\n<li>LlamaIndex, now in <a href=\"https:\/\/blog.llamaindex.ai\/announcing-llamaindex-0-9-719f03282945\">version 0.9<\/a>, also released many improvements, including <a href=\"https:\/\/llamahub.ai\/?tab=llama_packs\"><strong>LlaMa Packs<\/strong><\/a> featuring <strong>community modules<\/strong> such as RAG templates, <a href=\"https:\/\/blog.llamaindex.ai\/introducing-llama-datasets-aadb9994ad9e\"><strong>Llama Datasets<\/strong><\/a> for <strong>benchmarking<\/strong> RAG applications, <a href=\"https:\/\/blog.llamaindex.ai\/introducing-query-pipelines-025dc2bb0537\"><strong>Query Pipelines<\/strong><\/a> for improved workflow <strong>orchestration<\/strong>, and new custom and multimodal <a href=\"https:\/\/docs.llamaindex.ai\/en\/latest\/examples\/agent\/custom_agent.html\">ReAct agents<\/a>.<\/li>\n<\/ul>\n\n\n\n<h2 class=\"wp-block-heading\" id=\"Business\">Business Soap Opera<\/h2>\n\n\n\n<ul class=\"wp-block-list\">\n<li>OpenAI, Apple, and Google are trying to make <a href=\"https:\/\/www.theinformation.com\/articles\/openai-offers-publishers-as-little-as-1-million-a-year\"><strong>deals with publishers<\/strong><\/a> to use their content, with <strong>OpenAI<\/strong> successfully <a href=\"https:\/\/www.axelspringer.com\/en\/ax-press-release\/axel-springer-and-openai-partner-to-deepen-beneficial-use-of-ai-in-journalism\"><strong>partnering with Axel Springer<\/strong><\/a>.<\/li>\n\n\n\n<li>However, <strong>The<\/strong> <strong>New York Times<\/strong> rejected a retrospective partnership with <strong>OpenAI<\/strong> and initiated a <a href=\"https:\/\/www.theverge.com\/2023\/12\/27\/24016212\/new-york-times-openai-microsoft-lawsuit-copyright-infringement\"><strong>prominent lawsuit<\/strong><\/a>. They accused OpenAI of unlawful <strong>data use in training<\/strong> AI models and <strong>reputational damage<\/strong> to Wirecutter reviews through LLM hallucinations. <a href=\"https:\/\/openai.com\/blog\/openai-and-journalism\">OpenAI&#8217;s response<\/a> emphasized <strong>fair use<\/strong> in training, provided opt-in, and attributed <strong>regurgitation<\/strong> to a <strong>rare bug<\/strong> cherry-picked with very specific prompting. <a href=\"https:\/\/www.linkedin.com\/in\/gary-marcus-b6384b4\/\">Gary Marcus<\/a> <a href=\"https:\/\/garymarcus.substack.com\/p\/the-desperate-race-to-save-generative\">highlights<\/a> the LLM&#8217;s <strong>inherent<\/strong> tendency to <strong>regurgitate<\/strong> bits of text and argues against OpenAI&#8217;s <strong>resistance to paying licensing fees<\/strong> (while they already license some data use!). <a href=\"https:\/\/www.linkedin.com\/in\/andrewyng\/\">Andrew Ng<\/a> <a href=\"https:\/\/www.deeplearning.ai\/the-batch\/issue-230\/\">defends<\/a> OpenAI, as he views reading documents as fair use, and suggests that <strong>regurgitation may come from RAG<\/strong> rather than training. Gary Marcus disputes fair use, dismisses RAG as a <strong>red herring<\/strong>, and points out that Ng&#8217;s involvement in Gen AI companies influences his perspectives. Meanwhile, <strong>Japan<\/strong> has already <a href=\"https:\/\/www.biia.com\/japan-goes-all-in-copyright-doesnt-apply-to-ai-training\/\"><strong>removed AI training from copyright<\/strong><\/a>.<\/li>\n\n\n\n<li><strong>ByteDance<\/strong> (TikTok) violated OpenAI terms of service by <strong>training<\/strong> its own <strong>competing<\/strong> <strong>model<\/strong> (project Seed) <strong>using GPT-4<\/strong> and was <a href=\"https:\/\/the-decoder.com\/openai-bans-tiktok-company-bytedance-from-chatgpt-due-to-possible-data-theft\/\"><strong>banned from OpenAI<\/strong><\/a>.<\/li>\n\n\n\n<li>OpenAI activated the <a href=\"https:\/\/openai.com\/blog\/introducing-the-gpt-store\"><strong>GPTs store<\/strong><\/a>, with numerous GPT&#8217;s emerging. The trendy ones are often duplicated by others, possibly due to the ease of <strong>extracting custom prompts<\/strong> through <a href=\"https:\/\/andrew-horton.medium.com\/a-universal-prompt-injection-attack-in-the-gpt-store-6cacf6d887c0\">prompt injection attacks<\/a>. This emphasizes the <strong>importance of the knowledge files<\/strong> in creating a unique GPT. The leaderboard highlights <strong>Consensus<\/strong>, a research assistant, as the <strong>most popular<\/strong> GPT, suggesting that researchers and students are still the primary users of the technology.<\/li>\n\n\n\n<li>Following <strong>Nvidia<\/strong>&#8216;s unveiling of the <strong>H200 AI chip<\/strong> (see <a href=\"https:\/\/geneea.com\/news\/geneeas-ai-spotlight-7\/#media\">AI Spotlight #7<\/a>), <strong>Intel<\/strong> swiftly countered with their <a href=\"https:\/\/www.cnbc.com\/2023\/12\/14\/intel-unveils-gaudi3-ai-chip-to-compete-with-nvidia-and-amd.html\"><strong>Gaudi3 chip<\/strong><\/a>, and <strong>AMD<\/strong> is in the mix with <a href=\"https:\/\/www.ft.com\/content\/fa0c97af-c20f-461e-96c9-f2357496c599\"><strong>MI300X<\/strong><\/a> &#8211; all set to hit the market in the first quarter of this year. Meanwhile, a newcomer, Etched<strong>,<\/strong> <a href=\"https:\/\/www.youtube.com\/watch?v=K0XZ_ShxWkI&amp;t=2m11s\">aims<\/a> to outpace the giants by building <strong>transformer architecture<\/strong> directly <strong>into<\/strong> their <strong>chip<\/strong> for even faster inference.<\/li>\n<\/ul>\n\n\n\n<p>Please <a href=\"https:\/\/www.linkedin.com\/pulse\/geneeas-ai-spotlight-8-geneea-wwrie\/\">subscribe<\/a> and stay tuned for the next issue of Geneea\u2019s AI Spotlight newsletter!<\/p>\n","protected":false},"excerpt":{"rendered":"<p>The eighth edition of our newsletter on Large Language Models is out!<\/p>\n<p>This time, we look at<\/p>\n<p>\u25aa the future of AI with Altman and Nadella,<br \/>\n\u25aa some of the challenges that still separate us from the future Altman and Nadella describe,<br \/>\n\u25aa the spread of smaller models and new compression techniques,<br \/>\n\u25aa library updates, and finally,<br \/>\n\u25aa some corporate clashes.<\/p>\n","protected":false},"author":15,"featured_media":1731,"comment_status":"closed","ping_status":"closed","sticky":false,"template":"","format":"standard","meta":{"inline_featured_image":false,"footnotes":""},"categories":[378,374],"tags":[244,240,242],"class_list":["post-1730","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-large-language-models","category-newsletter","tag-ai","tag-generativeai","tag-newsletter"],"yoast_head":"<!-- This site is optimized with the Yoast SEO plugin v27.3 - https:\/\/yoast.com\/product\/yoast-seo-wordpress\/ -->\n<title>Geneea&#039;s AI Spotlight #8 - Geneea News<\/title>\n<meta name=\"description\" content=\"LLM newsletter #8: the future of AI with Altman and Nadella, spread of smaller models and new compression techniques, library updates, corporate clashes.\" \/>\n<meta name=\"robots\" content=\"index, follow, max-snippet:-1, max-image-preview:large, max-video-preview:-1\" \/>\n<link rel=\"canonical\" href=\"https:\/\/geneea.com\/news\/geneeas-ai-spotlight-8\" \/>\n<meta property=\"og:locale\" content=\"en_US\" \/>\n<meta property=\"og:type\" content=\"article\" \/>\n<meta property=\"og:title\" content=\"Geneea&#039;s AI Spotlight #8 - Geneea News\" \/>\n<meta property=\"og:description\" content=\"LLM newsletter #8: the future of AI with Altman and Nadella, spread of smaller models and new compression techniques, library updates, corporate clashes.\" \/>\n<meta property=\"og:url\" content=\"https:\/\/geneea.com\/news\/geneeas-ai-spotlight-8\" \/>\n<meta property=\"og:site_name\" content=\"Geneea News\" \/>\n<meta property=\"article:published_time\" content=\"2024-02-06T10:22:57+00:00\" \/>\n<meta property=\"article:modified_time\" content=\"2026-01-27T20:54:13+00:00\" \/>\n<meta property=\"og:image\" content=\"https:\/\/geneea.com\/news\/wp-content\/uploads\/2024\/02\/newsletter-8-robot.png\" \/>\n\t<meta property=\"og:image:width\" content=\"1280\" \/>\n\t<meta property=\"og:image:height\" content=\"719\" \/>\n\t<meta property=\"og:image:type\" content=\"image\/png\" \/>\n<meta name=\"author\" content=\"Marcela Soukupova\" \/>\n<meta name=\"twitter:card\" content=\"summary_large_image\" \/>\n<meta name=\"twitter:label1\" content=\"Written by\" \/>\n\t<meta name=\"twitter:data1\" content=\"Marcela Soukupova\" \/>\n\t<meta name=\"twitter:label2\" content=\"Est. reading time\" \/>\n\t<meta name=\"twitter:data2\" content=\"8 minutes\" \/>\n<script type=\"application\/ld+json\" class=\"yoast-schema-graph\">{\"@context\":\"https:\\\/\\\/schema.org\",\"@graph\":[{\"@type\":\"Article\",\"@id\":\"https:\\\/\\\/geneea.com\\\/news\\\/geneeas-ai-spotlight-8#article\",\"isPartOf\":{\"@id\":\"https:\\\/\\\/geneea.com\\\/news\\\/geneeas-ai-spotlight-8\"},\"author\":{\"name\":\"Marcela Soukupova\",\"@id\":\"https:\\\/\\\/geneea.com\\\/news\\\/#\\\/schema\\\/person\\\/69c8751a4c026723f4bac2e892f52cd8\"},\"headline\":\"Geneea&#8217;s AI Spotlight #8\",\"datePublished\":\"2024-02-06T10:22:57+00:00\",\"dateModified\":\"2026-01-27T20:54:13+00:00\",\"mainEntityOfPage\":{\"@id\":\"https:\\\/\\\/geneea.com\\\/news\\\/geneeas-ai-spotlight-8\"},\"wordCount\":1589,\"publisher\":{\"@id\":\"https:\\\/\\\/geneea.com\\\/news\\\/#organization\"},\"image\":{\"@id\":\"https:\\\/\\\/geneea.com\\\/news\\\/geneeas-ai-spotlight-8#primaryimage\"},\"thumbnailUrl\":\"https:\\\/\\\/geneea.com\\\/news\\\/wp-content\\\/uploads\\\/2024\\\/02\\\/newsletter-8-robot.png\",\"keywords\":[\"AI\",\"generativeAI\",\"newsletter\"],\"articleSection\":[\"Large language models\",\"Newsletter\"],\"inLanguage\":\"en-US\"},{\"@type\":\"WebPage\",\"@id\":\"https:\\\/\\\/geneea.com\\\/news\\\/geneeas-ai-spotlight-8\",\"url\":\"https:\\\/\\\/geneea.com\\\/news\\\/geneeas-ai-spotlight-8\",\"name\":\"Geneea's AI Spotlight #8 - Geneea News\",\"isPartOf\":{\"@id\":\"https:\\\/\\\/geneea.com\\\/news\\\/#website\"},\"primaryImageOfPage\":{\"@id\":\"https:\\\/\\\/geneea.com\\\/news\\\/geneeas-ai-spotlight-8#primaryimage\"},\"image\":{\"@id\":\"https:\\\/\\\/geneea.com\\\/news\\\/geneeas-ai-spotlight-8#primaryimage\"},\"thumbnailUrl\":\"https:\\\/\\\/geneea.com\\\/news\\\/wp-content\\\/uploads\\\/2024\\\/02\\\/newsletter-8-robot.png\",\"datePublished\":\"2024-02-06T10:22:57+00:00\",\"dateModified\":\"2026-01-27T20:54:13+00:00\",\"description\":\"LLM newsletter #8: the future of AI with Altman and Nadella, spread of smaller models and new compression techniques, library updates, corporate clashes.\",\"breadcrumb\":{\"@id\":\"https:\\\/\\\/geneea.com\\\/news\\\/geneeas-ai-spotlight-8#breadcrumb\"},\"inLanguage\":\"en-US\",\"potentialAction\":[{\"@type\":\"ReadAction\",\"target\":[\"https:\\\/\\\/geneea.com\\\/news\\\/geneeas-ai-spotlight-8\"]}]},{\"@type\":\"ImageObject\",\"inLanguage\":\"en-US\",\"@id\":\"https:\\\/\\\/geneea.com\\\/news\\\/geneeas-ai-spotlight-8#primaryimage\",\"url\":\"https:\\\/\\\/geneea.com\\\/news\\\/wp-content\\\/uploads\\\/2024\\\/02\\\/newsletter-8-robot.png\",\"contentUrl\":\"https:\\\/\\\/geneea.com\\\/news\\\/wp-content\\\/uploads\\\/2024\\\/02\\\/newsletter-8-robot.png\",\"width\":1280,\"height\":719},{\"@type\":\"BreadcrumbList\",\"@id\":\"https:\\\/\\\/geneea.com\\\/news\\\/geneeas-ai-spotlight-8#breadcrumb\",\"itemListElement\":[{\"@type\":\"ListItem\",\"position\":1,\"name\":\"Home\",\"item\":\"https:\\\/\\\/geneea.com\\\/news\"},{\"@type\":\"ListItem\",\"position\":2,\"name\":\"Geneea&#8217;s AI Spotlight #8\"}]},{\"@type\":\"WebSite\",\"@id\":\"https:\\\/\\\/geneea.com\\\/news\\\/#website\",\"url\":\"https:\\\/\\\/geneea.com\\\/news\\\/\",\"name\":\"Geneea News\",\"description\":\"Learn more about what&#039;s happening at Geneea: new NLP features, newest case studies, tutoring projects, conferences we attended, etc.\",\"publisher\":{\"@id\":\"https:\\\/\\\/geneea.com\\\/news\\\/#organization\"},\"potentialAction\":[{\"@type\":\"SearchAction\",\"target\":{\"@type\":\"EntryPoint\",\"urlTemplate\":\"https:\\\/\\\/geneea.com\\\/news\\\/?s={search_term_string}\"},\"query-input\":{\"@type\":\"PropertyValueSpecification\",\"valueRequired\":true,\"valueName\":\"search_term_string\"}}],\"inLanguage\":\"en-US\"},{\"@type\":\"Organization\",\"@id\":\"https:\\\/\\\/geneea.com\\\/news\\\/#organization\",\"name\":\"Geneea News\",\"url\":\"https:\\\/\\\/geneea.com\\\/news\\\/\",\"logo\":{\"@type\":\"ImageObject\",\"inLanguage\":\"en-US\",\"@id\":\"https:\\\/\\\/geneea.com\\\/news\\\/#\\\/schema\\\/logo\\\/image\\\/\",\"url\":\"https:\\\/\\\/geneea.com\\\/news\\\/wp-content\\\/uploads\\\/2022\\\/02\\\/cropped-geneea-logo-50pc.png\",\"contentUrl\":\"https:\\\/\\\/geneea.com\\\/news\\\/wp-content\\\/uploads\\\/2022\\\/02\\\/cropped-geneea-logo-50pc.png\",\"width\":242,\"height\":64,\"caption\":\"Geneea News\"},\"image\":{\"@id\":\"https:\\\/\\\/geneea.com\\\/news\\\/#\\\/schema\\\/logo\\\/image\\\/\"}},{\"@type\":\"Person\",\"@id\":\"https:\\\/\\\/geneea.com\\\/news\\\/#\\\/schema\\\/person\\\/69c8751a4c026723f4bac2e892f52cd8\",\"name\":\"Marcela Soukupova\",\"image\":{\"@type\":\"ImageObject\",\"inLanguage\":\"en-US\",\"@id\":\"https:\\\/\\\/secure.gravatar.com\\\/avatar\\\/44f35824640c6a5b31bfef2f478d704874dc3d81bfad511c158ab12274072e16?s=96&d=mm&r=g\",\"url\":\"https:\\\/\\\/secure.gravatar.com\\\/avatar\\\/44f35824640c6a5b31bfef2f478d704874dc3d81bfad511c158ab12274072e16?s=96&d=mm&r=g\",\"contentUrl\":\"https:\\\/\\\/secure.gravatar.com\\\/avatar\\\/44f35824640c6a5b31bfef2f478d704874dc3d81bfad511c158ab12274072e16?s=96&d=mm&r=g\",\"caption\":\"Marcela Soukupova\"},\"sameAs\":[\"http:\\\/\\\/Marcela%20Soukupova\"],\"url\":\"https:\\\/\\\/geneea.com\\\/news\\\/author\\\/marcela-soukupova\"}]}<\/script>\n<!-- \/ Yoast SEO plugin. -->","yoast_head_json":{"title":"Geneea's AI Spotlight #8 - Geneea News","description":"LLM newsletter #8: the future of AI with Altman and Nadella, spread of smaller models and new compression techniques, library updates, corporate clashes.","robots":{"index":"index","follow":"follow","max-snippet":"max-snippet:-1","max-image-preview":"max-image-preview:large","max-video-preview":"max-video-preview:-1"},"canonical":"https:\/\/geneea.com\/news\/geneeas-ai-spotlight-8","og_locale":"en_US","og_type":"article","og_title":"Geneea's AI Spotlight #8 - Geneea News","og_description":"LLM newsletter #8: the future of AI with Altman and Nadella, spread of smaller models and new compression techniques, library updates, corporate clashes.","og_url":"https:\/\/geneea.com\/news\/geneeas-ai-spotlight-8","og_site_name":"Geneea News","article_published_time":"2024-02-06T10:22:57+00:00","article_modified_time":"2026-01-27T20:54:13+00:00","og_image":[{"width":1280,"height":719,"url":"https:\/\/geneea.com\/news\/wp-content\/uploads\/2024\/02\/newsletter-8-robot.png","type":"image\/png"}],"author":"Marcela Soukupova","twitter_card":"summary_large_image","twitter_misc":{"Written by":"Marcela Soukupova","Est. reading time":"8 minutes"},"schema":{"@context":"https:\/\/schema.org","@graph":[{"@type":"Article","@id":"https:\/\/geneea.com\/news\/geneeas-ai-spotlight-8#article","isPartOf":{"@id":"https:\/\/geneea.com\/news\/geneeas-ai-spotlight-8"},"author":{"name":"Marcela Soukupova","@id":"https:\/\/geneea.com\/news\/#\/schema\/person\/69c8751a4c026723f4bac2e892f52cd8"},"headline":"Geneea&#8217;s AI Spotlight #8","datePublished":"2024-02-06T10:22:57+00:00","dateModified":"2026-01-27T20:54:13+00:00","mainEntityOfPage":{"@id":"https:\/\/geneea.com\/news\/geneeas-ai-spotlight-8"},"wordCount":1589,"publisher":{"@id":"https:\/\/geneea.com\/news\/#organization"},"image":{"@id":"https:\/\/geneea.com\/news\/geneeas-ai-spotlight-8#primaryimage"},"thumbnailUrl":"https:\/\/geneea.com\/news\/wp-content\/uploads\/2024\/02\/newsletter-8-robot.png","keywords":["AI","generativeAI","newsletter"],"articleSection":["Large language models","Newsletter"],"inLanguage":"en-US"},{"@type":"WebPage","@id":"https:\/\/geneea.com\/news\/geneeas-ai-spotlight-8","url":"https:\/\/geneea.com\/news\/geneeas-ai-spotlight-8","name":"Geneea's AI Spotlight #8 - Geneea News","isPartOf":{"@id":"https:\/\/geneea.com\/news\/#website"},"primaryImageOfPage":{"@id":"https:\/\/geneea.com\/news\/geneeas-ai-spotlight-8#primaryimage"},"image":{"@id":"https:\/\/geneea.com\/news\/geneeas-ai-spotlight-8#primaryimage"},"thumbnailUrl":"https:\/\/geneea.com\/news\/wp-content\/uploads\/2024\/02\/newsletter-8-robot.png","datePublished":"2024-02-06T10:22:57+00:00","dateModified":"2026-01-27T20:54:13+00:00","description":"LLM newsletter #8: the future of AI with Altman and Nadella, spread of smaller models and new compression techniques, library updates, corporate clashes.","breadcrumb":{"@id":"https:\/\/geneea.com\/news\/geneeas-ai-spotlight-8#breadcrumb"},"inLanguage":"en-US","potentialAction":[{"@type":"ReadAction","target":["https:\/\/geneea.com\/news\/geneeas-ai-spotlight-8"]}]},{"@type":"ImageObject","inLanguage":"en-US","@id":"https:\/\/geneea.com\/news\/geneeas-ai-spotlight-8#primaryimage","url":"https:\/\/geneea.com\/news\/wp-content\/uploads\/2024\/02\/newsletter-8-robot.png","contentUrl":"https:\/\/geneea.com\/news\/wp-content\/uploads\/2024\/02\/newsletter-8-robot.png","width":1280,"height":719},{"@type":"BreadcrumbList","@id":"https:\/\/geneea.com\/news\/geneeas-ai-spotlight-8#breadcrumb","itemListElement":[{"@type":"ListItem","position":1,"name":"Home","item":"https:\/\/geneea.com\/news"},{"@type":"ListItem","position":2,"name":"Geneea&#8217;s AI Spotlight #8"}]},{"@type":"WebSite","@id":"https:\/\/geneea.com\/news\/#website","url":"https:\/\/geneea.com\/news\/","name":"Geneea News","description":"Learn more about what&#039;s happening at Geneea: new NLP features, newest case studies, tutoring projects, conferences we attended, etc.","publisher":{"@id":"https:\/\/geneea.com\/news\/#organization"},"potentialAction":[{"@type":"SearchAction","target":{"@type":"EntryPoint","urlTemplate":"https:\/\/geneea.com\/news\/?s={search_term_string}"},"query-input":{"@type":"PropertyValueSpecification","valueRequired":true,"valueName":"search_term_string"}}],"inLanguage":"en-US"},{"@type":"Organization","@id":"https:\/\/geneea.com\/news\/#organization","name":"Geneea News","url":"https:\/\/geneea.com\/news\/","logo":{"@type":"ImageObject","inLanguage":"en-US","@id":"https:\/\/geneea.com\/news\/#\/schema\/logo\/image\/","url":"https:\/\/geneea.com\/news\/wp-content\/uploads\/2022\/02\/cropped-geneea-logo-50pc.png","contentUrl":"https:\/\/geneea.com\/news\/wp-content\/uploads\/2022\/02\/cropped-geneea-logo-50pc.png","width":242,"height":64,"caption":"Geneea News"},"image":{"@id":"https:\/\/geneea.com\/news\/#\/schema\/logo\/image\/"}},{"@type":"Person","@id":"https:\/\/geneea.com\/news\/#\/schema\/person\/69c8751a4c026723f4bac2e892f52cd8","name":"Marcela Soukupova","image":{"@type":"ImageObject","inLanguage":"en-US","@id":"https:\/\/secure.gravatar.com\/avatar\/44f35824640c6a5b31bfef2f478d704874dc3d81bfad511c158ab12274072e16?s=96&d=mm&r=g","url":"https:\/\/secure.gravatar.com\/avatar\/44f35824640c6a5b31bfef2f478d704874dc3d81bfad511c158ab12274072e16?s=96&d=mm&r=g","contentUrl":"https:\/\/secure.gravatar.com\/avatar\/44f35824640c6a5b31bfef2f478d704874dc3d81bfad511c158ab12274072e16?s=96&d=mm&r=g","caption":"Marcela Soukupova"},"sameAs":["http:\/\/Marcela%20Soukupova"],"url":"https:\/\/geneea.com\/news\/author\/marcela-soukupova"}]}},"_links":{"self":[{"href":"https:\/\/geneea.com\/news\/wp-json\/wp\/v2\/posts\/1730","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/geneea.com\/news\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/geneea.com\/news\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/geneea.com\/news\/wp-json\/wp\/v2\/users\/15"}],"replies":[{"embeddable":true,"href":"https:\/\/geneea.com\/news\/wp-json\/wp\/v2\/comments?post=1730"}],"version-history":[{"count":3,"href":"https:\/\/geneea.com\/news\/wp-json\/wp\/v2\/posts\/1730\/revisions"}],"predecessor-version":[{"id":1833,"href":"https:\/\/geneea.com\/news\/wp-json\/wp\/v2\/posts\/1730\/revisions\/1833"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/geneea.com\/news\/wp-json\/wp\/v2\/media\/1731"}],"wp:attachment":[{"href":"https:\/\/geneea.com\/news\/wp-json\/wp\/v2\/media?parent=1730"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/geneea.com\/news\/wp-json\/wp\/v2\/categories?post=1730"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/geneea.com\/news\/wp-json\/wp\/v2\/tags?post=1730"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}