Sebastian Berns, a doctoral researcher at Queen Mary University of London, is a big proponent of this crooked feature of chatbots which others abhor. He likes to use these chatbots because they hallucinate, turning them into valuable “co-creative partners”.
There’s an interesting part in ChatGPT. We can increase the hallucination by raising the temperature, making it more creative. The temperature here is not the usual thermodynamics one, but it is a hyper-parameter used to modulate the randomness and creativity of generated text.
For instance, while giving prompts to ChatGPT, users can also set temperatures between 0 and 100 and witness the highly creative response of the chatbot. At 100, the response becomes extremely random and unpredictable, often losing coherence and meaning. Here's an example: "Happiness, banana spaceship jellybean hippopotamus frolicking on a polka-dotted rainbow while playing a saxophone made of marshmallows and sprinkles."
Experts think that the word hallucination is not rightly aligned with chatbots. It should be called something else. In the latest edition of the ‘Schizophrenia Bulletin’, Oxford researchers published a piece titled, ‘False Responses From Artificial Intelligence Models Are Not Hallucinations’.
The paper highlights a potential concern regarding "hallucinations" in chatbots, which could pose a greater risk as these AI systems become more accurate. The danger lies in users overly depending on chatbots, particularly in their respective fields of expertise, due to the chatbots consistently offering accurate information.
Read the full story here.
Putting Common Sense in Machines
Former Google DeepMind senior research scientist Tejas Kulkarni, along with Max Kleiman-Weiner and scout investor at Sequoia Capital, embarked on a quest to harness the elusive 'common sense' to advance artificial general intelligence (AGI). Their endeavor, Common Sense Machines (CSM), introduces an end-to-end generative AI platform dedicated to crafting game-engine-ready content swiftly.
Kulkarni believes 3D has been a longstanding enigma in AI, prompting CSM to develop assets for the metaverse and gaming industry. Their user-generated content (UGC) approach aims to simplify user experiences, appealing to a broad audience, from hobbyists to 3D artists and studios. CSM's groundbreaking innovation, CSM Cube explores diverse input modalities like text, images, and videos, incorporating geometric deep learning, neural radiance fields, and 3D computer vision.
Read the full story here.
How to Get Hired at Zoho
Indian SaaS unicorn Zoho has fully integrated AI into its operations, leveraging in-house technology infrastructure for data science and AI tools. The company is actively hiring in the AI and ML domain, seeking professionals experienced in algorithms, machine learning, and handling large datasets. These roles encompass developing AI and ML applications, optimizing models, and staying current with AI advancements.
Zoho's AI research team, a crucial element within the company's hybrid AI structure, works closely with its central AI team and product-specific experts to implement tailored solutions. The hiring process values technical skills, problem-solving abilities, and alignment with company values, with an emphasis on programming proficiency in Java, Python, and C.
Read the full story here.
Green Apple
Комментариев нет:
Отправить комментарий
Примечание. Отправлять комментарии могут только участники этого блога.