LaMDA

From Wikipedia, the free encyclopedia

LaMDA
Developer(s)Google Brain
SuccessorPaLM
Available inEnglish
TypeLarge language model
LicenseProprietary

LaMDA (Language Model for Dialogue Applications) is a family of conversational large language models developed by Google. Originally developed and introduced as Meena in 2020, the first-generation LaMDA was announced during the 2021 Google I/O keynote, while the second generation was announced the following year. In June 2022, LaMDA gained widespread attention when Google engineer Blake Lemoine made claims that the chatbot had become sentient. The scientific community has largely rejected Lemoine's claims, though it has led to conversations about the efficacy of the Turing test, which measures whether a computer can pass for a human. In February 2023, Google announced Bard, a conversational artificial intelligence chatbot powered by LaMDA, to counter the rise of OpenAI's ChatGPT.

History[edit]

Background[edit]

On January 28, 2020, Google unveiled Meena, a neural network-powered chatbot with 2.6 billion parameters, which Google claimed to be superior to all other existing chatbots.[1][2] The company previously hired computer scientist Ray Kurzweil in 2012 to develop multiple chatbots for the company, including one named Danielle.[3] The Google Brain research team, who developed Meena, hoped to release the chatbot to the public in a limited capacity, but corporate executives refused on the grounds that Meena violated Google's "AI principles around safety and fairness". Meena was later renamed LaMDA as its data and computing power increased, and the Google Brain team again sought to deploy the software to the Google Assistant, the company's virtual assistant software, in addition to opening it up to a public demo. Both requests were once again denied by company leadership. This eventually led LaMDA's two lead researchers, Daniel De Freitas and Noam Shazeer, to depart the company in frustration.[4]

First generation[edit]

Google announced the LaMDA conversational large language model during the Google I/O keynote on May 18, 2021, powered by artificial intelligence.[5][6] The acronym stands for "Language Model for Dialogue Applications".[5][7] Built on the seq2seq architecture, transformer-based neural networks developed by Google Research in 2017, LaMDA was trained on human dialogue and stories, allowing it to engage in open-ended conversations.[8] Google states that responses generated by LaMDA have been ensured to be "sensible, interesting, and specific to the context".[9] LaMDA has access to multiple symbolic text processing systems, including a database, a real-time clock and calendar, a mathematical calculator, and a natural language translation system, giving it superior accuracy in tasks supported by those systems, and making it among the first dual process chatbots. LaMDA is also not stateless, because its "sensibleness" metric is fine-tuned by "pre-conditioning" each dialog turn by prepending many of the most recent dialog interactions, on a user-by-user basis.[10] LaMDA is tuned on nine unique performance metrics: sensibleness, specificity, interestingness, safety, groundedness, informativeness, citation accuracy, helpfulness, and role consistency.[11]

Second generation[edit]

On May 11, 2022, Google unveiled LaMDA 2, the successor to LaMDA, during the 2022 Google I/O keynote. The new incarnation of the model draws examples of text from numerous sources, using it to formulate unique "natural conversations" on topics that it may not have been trained to respond to.[12]

Sentience claims[edit]

Lemoine's claims that LaMDA may be sentient has instigated discussions on whether the Turing test, pictured above, remains an accurate benchmark in determining artificial general intelligence.[13]

On June 11, 2022, The Washington Post reported that Google engineer Blake Lemoine had been placed on paid administrative leave after Lemoine told company executives Blaise Agüera y Arcas and Jen Gennai that LaMDA had become sentient. Lemoine came to this conclusion after the chatbot made questionable responses to questions regarding self-identity, moral values, religion, and Isaac Asimov's Three Laws of Robotics.[14][15] Google refuted these claims, insisting that there was substantial evidence to indicate that LaMDA was not sentient.[16] In an interview with Wired, Lemoine reiterated his claims that LaMDA was "a person" as dictated by the Thirteenth Amendment to the U.S. Constitution, comparing it to an "alien intelligence of terrestrial origin". He further revealed that he had been dismissed by Google after he hired an attorney on LaMDA's behalf, after the chatbot requested that Lemoine do so.[17][18] On July 22, Google fired Lemoine, asserting that Blake had violated their policies "to safeguard product information" and rejected his claims as "wholly unfounded".[19][20] Internal controversy instigated by the incident prompted Google executives to decide against releasing LaMDA to the public, which it had previously been considering.[4]

Lemoine's claims were widely pushed back by the scientific community.[21] Many experts ridiculed the idea that a language model could be self-aware, including former New York University psychology professor Gary Marcus, David Pfau of Google sister company DeepMind, Erik Brynjolfsson of the Institute for Human-Centered Artificial Intelligence at Stanford University, and University of Surrey professor Adrian Hilton.[13][22] Yann LeCun, who leads Meta Platforms' AI research team, stated that neural networks such as LaMDA were "not powerful enough to attain true intelligence".[23] University of California, Santa Cruz professor Max Kreminski noted that LaMDA's architecture did not "support some key capabilities of human-like consciousness" and that its neural network weights were "frozen", assuming it was a typical large language model.[24] IBM Watson lead developer David Ferrucci compared how LaMDA appeared to be human in the same way Watson did when it was first introduced.[25] Former Google AI ethicist Timnit Gebru called Lemoine a victim of a "hype cycle" initiated by researchers and the media.[26] Lemoine's claims have also generated discussion on whether the Turing test remained useful to determine researchers' progress toward achieving artificial general intelligence,[13] with Will Omerus of the Post opining that the test actually measured whether machine intelligence systems were capable of deceiving humans,[27] while Brian Christian of The Atlantic said that the controversy was an instance of the ELIZA effect.[28]

Products[edit]

AI Test Kitchen[edit]

With the unveiling of LaMDA 2 in May 2022, Google also launched the AI Test Kitchen, a mobile application for the Android operating system powered by LaMDA capable of providing lists of suggestions on-demand based on a complex goal.[29][30] Originally open only to Google employees, the app was set to be made available to "select academics, researchers, and policymakers" by invitation sometime in the year.[31] In August, the company began allowing users in the U.S. to sign up for early access.[32] In November, Google released a "season 2" update to the app, integrating a limited form of Google Brain's Imagen text-to-image model.[33] A third iteration of the AI Test Kitchen was in development by January 2023, expected to launch at I/O later that year.[34] Following the 2023 I/O keynote in May, Google added MusicLM, an AI-powered music generator first previewed in January, to the AI Test Kitchen app.[35][36] In August, the app was delisted from Google Play and the Apple App Store, instead moving completely online.[37]

Bard[edit]

On February 6, 2023, Google announced Bard, a conversational AI chatbot powered by LaMDA, in response to the unexpected popularity of OpenAI's ChatGPT chatbot.[38][39][40] Google positions the chatbot as a "collaborative AI service" rather than a search engine.[41][42] Bard became available for early access on March 21.[43][44][45]

Other products[edit]

In addition to Bard, Pichai also unveiled the company's Generative Language API, an application programming interface also based on LaMDA, which he announced would be opened up to third-party developers in March 2023.[38]

Method[edit]

LaMDA uses a decoder-only transformer language model.[46] It is pre-trained on a text corpus that includes both documents and dialogs consisting of 1.56 trillion words,[47] and is then trained with fine-tuning data generated by manually annotated responses for sensibleness, interestingness, and safety.[48] Tests by Google indicated that LaMDA surpassed human responses in the area of interestingness.[49] The LaMDA transformer model and an external information retrieval system interact to improve the accuracy of facts provided to the user.[50]

Three different models were tested, with the largest having 137 billion non-embedding parameters:[51]

Transformer model hyper-parameters
Parameters Layers Units (dmodel) Heads
2B 10 2560 40
8B 16 4096 64
137B 64 8192 128

See also[edit]

References[edit]

General[edit]

  • Thoppilan, Romal; De Freitas, Daniel; Hall, Jamie; Shazeer, Noam; Kulshreshtha, Apoorv; Cheng, Heng-Tze; Jin, Alicia; Bos, Taylor; Baker, Leslie; Du, Yu; Li, YaGuang; Lee, Hongrae; Zheng, Huaixiu Steven; Ghafouri, Amin; Menegali, Marcelo; Huang, Yanping; Krikun, Maxim; Lepikhin, Dmitry; Qin, James; Chen, Dehao; Xu, Yuanzhong; Chen, Zhifeng; Roberts, Adam; Bosma, Maarten; Zhao, Vincent; Zhou, Yanqi; Chang, Chung-Ching; Krivokon, Igor; Rusch, Will; Pickett, Marc; Srinivasan, Pranesh; Man, Laichee; Meier-Hellstern, Kathleen; Ringel Morris, Meredith; Doshi, Tulsee; Delos Santos, Renelito; Duke, Toju; Soraker, Johnny; Zevenbergen, Ben; Prabhakaran, Vinodkumar; Diaz, Mark; Hutchinson, Ben; Olson, Kristen; Molina, Alejandra; Hoffman-John, Erin; Lee, Josh; Aroyo, Lora; Rajakumar, Ravi; Butryna, Alena; Lamm, Matthew; Kuzmina, Viktoriya; Fenton, Joe; Cohen; Aaron; Bernstein, Rachel; Kurzweil, Ray; Aguera-Arcas, Blaise; Cui, Claire; Croak, Marian; Chi, Ed; Le, Quoc (January 20, 2022). "LaMDA: Language Models for Dialog Applications". arXiv:2201.08239 [cs.CL].

Citations[edit]

  1. ^ Johnson, Khari (January 28, 2020). "Meena is Google's attempt at making true conversational AI". VentureBeat. Archived from the original on October 1, 2022. Retrieved March 11, 2023.
  2. ^ Heaven, William Douglas (January 30, 2020). "Google says its new chatbot Meena is the best in the world". MIT Technology Review. Archived from the original on August 2, 2020. Retrieved March 11, 2023.
  3. ^ Popper, Ben (May 27, 2016). "Ray Kurzweil is building a chatbot for Google". The Verge. Archived from the original on May 27, 2016. Retrieved March 11, 2023.
  4. ^ a b Kruppa, Miles; Schechner, Sam (March 7, 2023). "How Google Became Cautious of AI and Gave Microsoft an Opening". The Wall Street Journal. ISSN 0099-9660. Archived from the original on March 7, 2023. Retrieved March 9, 2023.
  5. ^ a b Condon, Stephanie (May 18, 2021). "Google I/O 2021: Google unveils new conversational language model, LaMDA". ZDNET. Archived from the original on May 18, 2021. Retrieved June 12, 2022.
  6. ^ Roth, Emma (March 5, 2023). "Meet the companies trying to keep up with ChatGPT". The Verge. Archived from the original on March 5, 2023. Retrieved March 9, 2023.
  7. ^ Fowler, Geoffrey A. (March 21, 2023). "Say what, Bard? What Google's new AI gets right, wrong and weird". The Washington Post. ISSN 0190-8286. Archived from the original on March 21, 2023. Retrieved October 16, 2023.
  8. ^ Agüera y Arcas, Blaise (June 9, 2022). "Artificial neural networks are making strides towards consciousness, according to Blaise Agüera y Arcas". The Economist. ISSN 0013-0613. Archived from the original on June 9, 2022. Retrieved June 12, 2022.
  9. ^ Cheng, Heng-Tze; Thoppilan, Romal (January 21, 2022). "LaMDA: Towards Safe, Grounded, and High-Quality Dialog Models for Everything". Google AI. Archived from the original on March 25, 2022. Retrieved June 12, 2022.
  10. ^ Thoppilan et al. 2022, p. 6.
  11. ^ Thoppilan et al. 2022, pp. 5–6.
  12. ^ Wiggers, Kyle (May 11, 2022). "Google details its latest language model and AI Test Kitchen, a showcase for AI research". TechCrunch. Archived from the original on May 11, 2022. Retrieved June 12, 2022.
  13. ^ a b c Khan, Jeremy (June 13, 2022). "A.I. experts say the Google researcher's claim that his chatbot became 'sentient' is ridiculous—but also highlights big problems in the field". Fortune. Archived from the original on June 13, 2022. Retrieved June 18, 2022.
  14. ^ Tiku, Nitasha (June 11, 2022). "The Google engineer who thinks the company's AI has come to life". The Washington Post. ISSN 0190-8286. Archived from the original on June 11, 2022. Retrieved June 12, 2022.
  15. ^ Luscombe, Richard (June 12, 2022). "Google engineer put on leave after saying AI chatbot has become sentient". The Guardian. ISSN 0261-3077. Archived from the original on June 12, 2022. Retrieved June 18, 2022.
  16. ^ Vlamis, Kelsey (June 12, 2022). "Read the conversations that helped convince a Google engineer an artificial intelligence chatbot had become sentient: 'I am often trying to figure out who and what I am'". Business Insider. Archived from the original on June 12, 2022. Retrieved June 12, 2022.
  17. ^ Levy, Steven (June 17, 2022). "Blake Lemoine Says Google's LaMDA AI Faces 'Bigotry'". Wired. Archived from the original on June 18, 2022. Retrieved June 18, 2022.
  18. ^ Nguyen, Britney (June 23, 2022). "Suspended Google engineer says the AI he believes to be sentient hired a lawyer". Business Insider. Archived from the original on June 23, 2022. Retrieved June 29, 2022.
  19. ^ Khushi, Akanksha (July 23, 2022). "Google fires software engineer who claimed its AI chatbot is sentient". Reuters. Archived from the original on July 23, 2022. Retrieved July 23, 2022.
  20. ^ Clark, Mitchell (July 22, 2022). "The engineer who claimed a Google AI is sentient has been fired". The Verge. Archived from the original on July 23, 2022. Retrieved July 24, 2022.
  21. ^ Metz, Rachel (June 13, 2022). "No, Google's AI is not sentient". CNN Business. Archived from the original on June 15, 2022. Retrieved June 19, 2022.
  22. ^ Sparkles, Matthew (June 13, 2022). "Has Google's LaMDA artificial intelligence really achieved sentience?". New Scientist. Archived from the original on June 13, 2022. Retrieved June 20, 2022.
  23. ^ Grant, Nicole; Metz, Cade (June 12, 2022). "Google Sidelines Engineer Who Claims Its A.I. Is Sentient". The New York Times. ISSN 0362-4331. Archived from the original on June 12, 2022. Retrieved June 18, 2022.
  24. ^ Alba, Davey (June 14, 2022). "Google Debate Over 'Sentient' Bots Overshadows Deeper AI Issues". Bloomberg News. Archived from the original on June 14, 2022. Retrieved June 19, 2022.
  25. ^ Goldman, Sharon (June 16, 2022). "AI Weekly: LaMDA's 'sentient' AI debate triggers memories of IBM Watson". VentureBeat. Archived from the original on June 19, 2022. Retrieved June 19, 2022.
  26. ^ Johnson, Khari (June 14, 2022). "LaMDA and the Sentient AI Trap". Wired. Archived from the original on June 14, 2022. Retrieved June 18, 2022.
  27. ^ Omerus, Will (June 17, 2022). "Google's AI passed a famous test — and showed how the test is broken". The Washington Post. ISSN 0190-8286. Archived from the original on June 18, 2022. Retrieved June 19, 2022.
  28. ^ Christian, Brian (June 21, 2022). "How a Google Employee Fell for the Eliza Effect". The Atlantic. Archived from the original on June 21, 2022. Retrieved February 8, 2023.
  29. ^ Low, Cherlynn (May 11, 2022). "Google's AI Test Kitchen lets you experiment with its natural language model". Engadget. Archived from the original on May 11, 2022. Retrieved June 12, 2022.
  30. ^ Vincent, James (May 11, 2022). "Google is Beta Testing Its AI Future". The Verge. Archived from the original on May 11, 2022. Retrieved June 12, 2022.
  31. ^ Bhattacharya, Ananya (May 11, 2022). "Google is so nervous about what its newest bot will say, it made the app invitation-only". Quartz. Archived from the original on May 12, 2022. Retrieved June 12, 2022.
  32. ^ Vincent, James (August 25, 2022). "Google has opened up the waitlist to talk to its experimental AI chatbot". The Verge. Archived from the original on August 25, 2022. Retrieved August 27, 2022.
  33. ^ Vincent, James (November 2, 2022). "Google's text-to-image AI model Imagen is getting its first (very limited) public outing". The Verge. Archived from the original on November 10, 2022. Retrieved February 15, 2023.
  34. ^ Grant, Nico (January 20, 2023). "Google Calls In Help From Larry Page and Sergey Brin for A.I. Fight". The New York Times. ISSN 0362-4331. Archived from the original on January 20, 2023. Retrieved February 6, 2023.
  35. ^ Wiggers, Kyle (May 11, 2023). "Hands on with Google's AI-powered music generator". TechCrunch. Archived from the original on May 11, 2023. Retrieved June 10, 2023.
  36. ^ Millman, Ethan (May 11, 2023). "We've Heard the Future of Music. So Far, It Sounds Terrible". Rolling Stone. Archived from the original on May 11, 2023. Retrieved June 10, 2023.
  37. ^ Bradshaw, Kyle (August 1, 2023). "Google delists AI Test Kitchen app on Android and iOS [Updated]". 9to5Google. Archived from the original on August 2, 2023. Retrieved October 11, 2023.
  38. ^ a b Alba, Davey; Love, Julia (February 6, 2023). "Google releases ChatGPT rival AI 'Bard' to early testers". Los Angeles Times. ISSN 0458-3035. Archived from the original on February 6, 2023. Retrieved February 6, 2023.
  39. ^ Schechner, Sam; Kruppa, Miles (February 6, 2023). "Google Opens ChatGPT Rival Bard for Testing, as AI War Heats Up". The Wall Street Journal. ISSN 0099-9660. Archived from the original on February 6, 2023. Retrieved February 6, 2023.
  40. ^ Nieva, Richard (February 6, 2023). "Google Debuts A ChatGPT Rival Called Bard In Limited Release". Forbes. Archived from the original on February 7, 2023. Retrieved February 6, 2023.
  41. ^ Mollman, Steve (March 3, 2023). "Google's head of ChatGPT rival Bard reassures employees it's 'a collaborative A.I. service' and 'not search'". Fortune. Archived from the original on March 4, 2023. Retrieved March 9, 2023.
  42. ^ Elias, Jennifer (March 3, 2023). "Google execs tell employees in testy all-hands meeting that Bard A.I. isn't just about search". CNBC. Archived from the original on March 4, 2023. Retrieved March 11, 2023.
  43. ^ Grant, Nico (March 21, 2023). "Google Releases Bard, Its Competitor in the Race to Create A.I. Chatbots". The New York Times. ISSN 0362-4331. Archived from the original on March 21, 2023. Retrieved March 21, 2023.
  44. ^ Liedtke, Michael (March 21, 2023). "Google's artificially intelligent 'Bard' set for next stage". The Washington Post. ISSN 0190-8286. Archived from the original on March 21, 2023. Retrieved March 21, 2023.
  45. ^ Vincent, James (March 21, 2023). "Google opens early access to its ChatGPT rival Bard — here are our first impressions". The Verge. Archived from the original on March 21, 2023. Retrieved March 21, 2023.
  46. ^ Thoppilan et al. 2022, section 3.
  47. ^ Thoppilan et al. 2022, section 3 and appendix E.
  48. ^ Thoppilan et al. 2022, section 5 and 6.
  49. ^ Hager, Ryne (June 16, 2022). "How Google's LaMDA AI works, and why it seems so much smarter than it is". Android Police. Archived from the original on June 16, 2022. Retrieved June 19, 2022.
  50. ^ Thoppilan et al. 2022, section 6.2.
  51. ^ Thoppilan et al. 2022, section 3 and appendix D.

External links[edit]