Hi all(emaal),
Like always a lot is happening in the land of AI. This edition we’re seeing private schools further implementing AI, AI making volcano-scarred ancient texts readable, lots of government agencies setting up or (silently) killing AI centres, the first photography prize meant for AI generated imagery, big tech companies increasingly investing in making their own chips, and how Large Language Models can be used to help diagnose schizophrenia, amongst others. AI is nothing if not diverse.
On to the news-
The UK government has quietly dismissed the independent advisory board of its Centre for Data Ethics and Innovation (CDEI) — tasked with promoting the responsible deployment of data and AI technologies
USA's National Security Agency is setting up an AI security center
USA government announcing executive order, hinting at building out the earlier presented AI Bill of Rights
Chinese regulators creating a blacklist of sources they deem unsafe for AI models to train on, including content censored by the CCP
Google pledging its generative AI users that it will defend them if accused of intellectual property violations, joining Microsoft and Adobe with similar pledges made
Leaked prompt that ChatGPT uses for image generation with DALL-E
Renowned programming experts John Carmack and Richard Sutton teaming up to help establish, advance and document Artificial General Intelligence (AGI) signs of life.
USA's CIA integrating new AI tools to help sift through data
OpenAI's GPT4-V(ision) paper outlined several limitations of its new model capabilities
WallStreetJournal article on why AI tools are becoming ubiquitous in our digital lives, and suggesting an inevitability of their usage
Australian researchers designed an algorithm that can intercept malicious 'man in the middle' attacks on unmanned military robots with high accuracy
Bing chat fooled into solving Captcha anti-bot tests
DALL-E3 was published, showing performance (subjectively) seen as better than Midjourney, setting a new state of the art
Australian photography festival giving away first 'prompted peculiar international AI prize' AI-art prize
Wallstreet Journal asked its readers whether it matters if AI created art you enjoy, listing various responses they received
English private school implemented an 'AI headteacher' to support the school's headmaster in his tasks
Private school in USA using adaptive AI tools to 'optimize learning', to help students learn at their own pace at their own level
Integration of AI tools at Hong Kong University of Science and Technology (HKUST), stating AI will soon be scholars' "debate partner"
Diagnosing schizophrenia with the support of a Large Language Model
GPT4's medical soft skills compared to human medical practitioners, achieving similar results
New analysis by Dutch PhD student showcasing various scenarios of energy use by the 'AI boom', showing stellar numbers
Microsoft seemingly looking into nuclear energy as a clean source of energy for its energy hungry (AI) ambitions
Although CEO's push for integrating/exploring generative AI, CIO's see limited benefits - for now
Majority of job openings on generative AI topics in US concentrated around big tech, beating the narrative of 'democratisation of generative AI development'
New AI offerings currently not profitable for Big Tech companies
New AI tools may lead to new form of interface more well-suited to their abilities to see, hear, and speak to the user - and with it new forms of hardware separate from your phone
Intel fast approaching NVIDIA in cutting-edge hardware for running (a.o.) AI models
Microsoft developing their own in-house chips, soon to be published
OpenAI exploring development of their own in-house chips
(NL) AI richtlijnen binnen de journalistiek, hoe gaan Nederlandse redacties ermee om?
Nvidia and Foxconn teaming up to build 'AI factories', seemingly meaning a large server cluster that is performing various task for (re)training AI models to redistribute amongst users in a continuous fashion, targeting mainly EV's for now
Google agreeing to pay €3.2mil to German publishers for its publication of news content
Reka, startup founded by researchers from Deepmind, Meta AI, and Google, published their multi-modal LLM called Yasa-1 for businesses
Mistral unveiled their 7B parameter LLM that outperformes LLaMA2-13B, whilst being able to be comfortably put on a singular device/GPU, under a permissive Apache 2 license (commercial ends thus open)
GPT4-V unveiled, allowing GPT4 to hear and see, further enhancing it's potential
Baidu, a Chinese tech giant, unveiled their ERNIE 4.0 model, which the company tries to position as a viable alternative to OpenAI's ChatGPT
Some evidence that LLM's may actually have temporal-spatial representations of the world in their activations
Article showcasing through many experiments that 'emergent abilities' in LLMs are actually just in-context learning
Decoder-only auto-regressive multi-modal model, uses image retrieval augmentation for generating images, surprisingly good compared to present diffusion models
Survey of hallucination in LLMs, includes mitigation strategies and benchmarks you can use to measure it.
AI tool used to read ancient greek scroll that was charred by the Vesuvius
See you in the next one!
Leven is mooi
