The Dances of Behemoths: Microsoft AI Tie-Up Sparks FTC Probe

"The Federal Trade Commission is investigating whether Microsoft structured one of its latest deals with an artificial-intelligence startup to avoid a government antitrust review of the transaction.

Microsoft hired Inflection AI's co-founder and almost all of its employees in March and agreed to pay the startup around $650 million as part of a licensing fee to resell its technology. Inflection's investors were told they would be repaid over time by the sales proceeds.

Companies are required to report acquisitions valued at more than $119 million to federal antitrust-enforcement agencies, which have the option to investigate a deal's impact on competition. The FTC or the Justice Department can sue to block mergers or other investments if an investigation finds the deal would substantially reduce competition or lead to a monopoly.

A Microsoft spokeswoman denied any wrongdoing, and an Inflection official said he was unaware of the inquiry and that Microsoft didn't acquire his company.

The FTC is drilling down on Microsoft's deal with Inflection, seeking information about how and why they negotiated their partnership, according to a person familiar with the matter and records viewed by The Wall Street Journal. Civil subpoenas the commission sent recently to Microsoft and Inflection seek documents going back about two years. The agency is trying to determine whether Microsoft crafted a deal that would give it control of Inflection but also dodge FTC review of the transaction, the person said.

If the agency finds Microsoft should have reported and sought government review of its deal with Inflection, the FTC could bring an enforcement action against Microsoft. Officials could ask a court to fine Microsoft and suspend the transaction while the FTC conducts a full-scale investigation of the deal's impact on competition.

A Microsoft spokeswoman said the company is confident it complied with antitrust laws. Its agreements with Inflection gave Microsoft "the opportunity to recruit individuals at Inflection AI and build a team capable of accelerating Microsoft Copilot," the spokeswoman said, referring to the company's AI chatbot. Inflection continues as an independent business, she said.

Microsoft shares closed up slightly at $424.52 on Thursday.

The FTC has been sifting through AI investments made by leading companies such as Microsoft and Google-owner Alphabet. FTC Chair Lina Khan has expressed concern that technology behemoths could eventually acquire or control the most promising AI applications, giving them a tight grip on systems that have humanlike abilities to converse, create art and write computer code.

Tech companies often buy startups to scoop up their talent, a tactic known as an "acquihire." In Microsoft's case, the company picked off Inflection's specialized workforce of AI researchers but didn't purchase the company outright.

Based in the San Francisco Bay Area, Inflection AI built one of the world's biggest large language models and launched an AI chatbot with that technology called Pi. Inflection is one of the tech companies that built and sold access to large language models. The others include OpenAI, the creator of ChatGPT, and Google.

Microsoft was an investor in OpenAI and Inflection AI. The FTC opened an investigation in January of Microsoft's investment in OpenAI and Alphabet's relationship with Anthropic, a rival of OpenAI that was founded by former OpenAI engineers in 2021.

The FTC is close to finalizing an agreement with the Justice Department to divide responsibility for antitrust oversight of companies in the AI space, people familiar with the matter said. The FTC would have the ability to probe the practices of Microsoft and OpenAI, while the department would handle Nvidia, the semiconductor firm whose chips are essential to training AI models.

The deal between the agencies would allow the Justice Department to investigate whether any of Nvidia's chip-supply agreements or other practices are anticompetitive, one of the people said.

Nvidia declined to comment. The oversight arrangement was first reported by the New York Times.

At Microsoft, Inflection AI co-founder Mustafa Suleyman and his former team established a new division called Microsoft AI that was tasked with developing AI products for consumers. That includes AI assistants for Bing, its search engine, and Windows.

Inflection AI is continuing operations under a new management team but pivoted away from Pi, a consumer product, and toward services for corporate clients.

Ted Shelton, the company's new chief operating officer, said he wasn't aware of an FTC investigation. But Inflection wasn't acquired by Microsoft, he said. "We are a wholly independent company," Shelton said. "Microsoft has no investment in our company."

Entrepreneur Reid Hoffman and venture-capital firm Greylock Partners are now the principal investors in Inflection, Shelton said.

The hires resembled how Microsoft earlier moved to hire Sam Altman, the CEO of OpenAI, after his board of directors pushed him out in November. Altman returned as chief executive of OpenAI after a five-day standoff with the board. The directors alleged Altman wasn't completely candid in his communications with them.

OpenAI's nonprofit board oversees a for-profit arm that has outside investors. Microsoft invested around $13 billion in OpenAI, acquiring a claim to 49% of any profit it generates.

Wall Street Journal owner News Corp has a content-licensing partnership with OpenAI." [1]

1. Microsoft Tie-Up Sparks FTC Probe. Michaels, Dave; Dotan, Tom.  Wall Street Journal, Eastern edition; New York, N.Y.. 07 June 2024: A.1.

„Apple“ atsiliko dirbtinio intelekto arenoje, pasivyti yra tikslas

„Tiems, kurie jas matė, anksčiau šį dešimtmetį atnaujintos Siri demonstracijos Apple viduje parodė nuostabias galimybes, kurias gali turėti galingas dirbtinio intelekto (AI) balso asistentas.

 

 Garsusis asistentas, vienas iš paskutinių projektų, prie kurių „Apple“ įkūrėjas Steve'as Jobsas dirbo prieš mirtį, buvo visiškai perdarytas. Naujasis „Siri“, galintis veikti, naudojant „iPhone“ ir be interneto ryšio, sužavėjo žmones pagerintu greičiu, pokalbio galimybėmis ir tikslumu, kuriuo suprato vartotojo komandas. Kodiniu pavadinimu „Project Blackbird“ pastangos taip pat įsivaizdavo „Siri“ su galimybėmis, kurias sukūrė trečiųjų šalių programų kūrėjai, pasak žmonių, susipažinusių su šiuo darbu.

 

 Tačiau konkuruojantis projektas laimėjo vidinį konkursą prieš pat 10 metų nuo Siri paleidimo metines. Žinomas, kaip Siri X, kuklesnis atnaujinimas apėmė daugiau esamos Siri programinės įrangos perkėlimą į iPhone iš nuotolinių serverių, siekiant pagerinti balso asistento greitį ir privatumą. „Siri X“ patobulinimas buvo pristatytas 2021 m.

 

 Kitą savaitę „Apple“ kasmetinėje pasaulinėje kūrėjų konferencijoje bendrovė ketina prisijungti prie AI ginklavimosi lenktynių, kurios, daugelio nuomone, nulems technologijų ateitį. „iPhone“ gamintojas bando pasivyti „Microsoft“, „Alphabet“ „Google“ ir kitus konkurentus, kurie pradėjo integruoti generatyvųjį AI į savo pagrindinius produktus.

 

 „Apple“ atsargumas ir būdingas slaptumas, taip pat rūpestingumas, kurio ji imasi, atnaujindama įrenginius, kuriuose aparatinė ir programinė įranga yra sklandžiai integruota, sutrukdė jos ankstyvąsias pastangas dirbtinio intelekto arenoje, sakė žmonės. Dabar ji atsiduria neįprastoje padėtyje, kai turi rizikuoti.

 

 Bendrovė ketina paskelbti daugybę generatyvaus AI atnaujinimų jos programinės įrangos produktams, įskaitant „Siri“, sakė žmonės, susipažinę su jos planais. AI funkcijos apima pagalbą, rašant žinutes, redaguojant nuotraukas ir apibendrinant tekstus.

 

 Nors nesitikima, kad „Apple“ greitai pasivys pirmaujančius dirbtinio intelekto novatorius, bendrovė ruošiasi pristatyti AI funkcijas su įspūdingomis galimybėmis, kurios taip pat maksimaliai padidina vartotojų privatumą – tai bus pagrindinis rūpestis, norint išnaudoti visas AI padėjėjų galimybes.

 

 „Apple“ taip pat turėtų atskleisti vieną ar daugiau potencialių partnerysčių su pagrindiniais AI kūrėjais po derybų su „OpenAI“, „Google“ ir „Cohere“, sakė žmonės.

 

 „Apple“ jau seniai didžiuojasi savo gaminių pristatymo tobulumu, o tai beveik neįmanoma, naudojant naujus AI modelius. Nors OpenAI sistemos apakino daugiau, nei 180 milijonų vartotojų jų rašymo, vaizdų ir video įrašų gamyba, jos yra linkusios retkarčiais padaryti klaidų, dažnai vadinamų haliucinacijomis. „Apple“ turėjo ribotą toleranciją tokioms problemoms.

 

 „Su dirbtiniu intelektu nėra tokio dalyko, kaip 100% tikslumas, tai yra pagrindinė realybė“, – sakė Vašingtono universiteto kompiuterių mokslo ir inžinerijos profesorius emeritas Pedro Domingosas. ""Apple" su tuo nesuderinama. Jie nieko neišleis, kol nebus tobula."

 

 „Apple“ apsvarstė, ar leisti vartotojams pasirinkti trečiosios šalies AI teikėją, kuris galėtų pakeisti „Siri“ arba padėti jį valdyti, sakė vienas iš žmonių. Neaišku, kokiu būdu „Siri“ galėtų maitinti, papildyti ar pakeisti trečiųjų šalių AI teikėjai, ar „Apple“ imsis šios idėjos versijos.

 

 „Bloomberg“ anksčiau pranešė apie OpenAI partnerystę, o „Information“ pranešė apie pastangas pertvarkyti „Siri“.

 

 „Google“, „Microsoft“ ir „Samsung Electronics“ pradėjo sparčiai integruoti generatyvųjį AI į savo įrenginius ir paslaugas. Nors „Apple“ atsilieka nuo kartų kaitos technologijų pramonėje, daugelis investuotojų ir dirbtinio intelekto ekspertų teigė, kad bendrovė suras būdą, kaip generatyvųjį AI pritraukti į mases.

 

 „Apple“ gali padaryti beveik viską, ką tik užsimano“, – sakė Vineetas Khosla, buvęs „Siri“ komandos inžinierių vadovas, šiuo metu „Washington Post“ vyriausiasis technologijų pareigūnas. "Bendrovė yra orientuota į vartotojus. Jų AI tikslas yra užtikrinti, kad jis veiktų labai jautriai, atsižvelgiant į privatumą."

 

 Bėgant metams „Apple“ patobulino „Siri“ ir visuose savo produktuose įtraukė mažesnes dirbtinio intelekto funkcijas. Neseniai išleistose akiniuose „Vision Pro“ AI plačiai naudojamas akių ir rankų padėčiai sekti.

 

 Kai „Siri“ pasirodė 2011 m., „Apple“ aplenkė konkurentus, siekdama sukurti pirmąjį AI asistentą. Jobsas, vadovavęs įsigijimui 2010 m., po kurio atsirado Siri, paskatino komandą išlaikyti sausą asistento sąmojį ir humoro jausmą. Ankstyvas startas parodė įmonės norą rizikuoti.

 

 „Siri buvo paskutinis dalykas, kurį „Apple“ naudojo pirmaudama“, – sakė Dagas Kittlausas, „Apple“ nusipirkto „Siri“ startuolio įkūrėjas. Jis paliko „Apple“ netrukus po produkto pristatymo.

 

 Kadangi „Apple“ stengėsi toliau tobulinti „Siri“, bendrovė pasamdė vieną iš geriausių „Google“ inžinierių vadovų, kad vykdytų jos AI strategiją: Johną Giannandrea. 2018 metais jis buvo pakeltas į vyresniojo viceprezidento vaidmenį, tapo  tiesiogiai atskaitingas „Apple“ vadovui Timui Cookui.

 

 Ankstyvuose komandos susitikimuose Giannandrea aiškiai pasakė, kad pagrindinis dėmesys buvo skiriamas „Siri“ tobulinimui. Jam taip pat buvo pavesta centralizuoti didžiąją dalį suskaidytų „Apple“ pastangų, susijusių su AI. Jis pradėjo kurti savo dirbtinio intelekto komandą, įdarbindamas „Google“ darbuotojus ir įsigydamas startuolių, tačiau, pasak žmonių, susipažinusių su jų darbu, jo komandai buvo sunku prisitaikyti prie likusios „Apple“ grupės.

 

 Naujoji AI grupė veikė panašiai, kaip „Google“ dalys, kur terminai yra laisviau apibrėžti. „Apple“ komandos turi laikytis griežtų terminų, kad produktai būtų paruošti, kiekvieną rudenį išleistiems, renginiams. Pastangos bendradarbiauti tarp kitų „Apple“ dalių, kuriančių produktus, ir AI komandos kartais žlugo, nes nepavyko susitarti dėl terminų, sakė žmonės.

 

 Užuot dirbę su AI komanda, kitos Apple programinės įrangos produktų kūrimo dalys išlaikė savo atskiras AI galimybes. Pavyzdžiui, vyresniojo viceprezidento Craig Federighi vadovaujama programinės įrangos grupė toliau investavo ir kūrė AI, užtikrinantį vaizdo ir vaizdo atpažinimo galimybes, sakė buvę Apple darbuotojai.

 

 Kai 2022 m. pabaigoje pasirodė „ChatGPT“, viskas pasikeitė. Per Kalėdų pertrauką programinės įrangos vadovas Federighi tapo tikinčiu po to, kai pradėjo žaisti su „Microsoft“ priklausančiu „GitHub AI“ kodavimo įrankiu „Copilot“, kuris yra pagrįstas OpenAI technologija, sakė su jo patirtimi susipažinę žmonės.

 

 „Apple“ sustiprino pastangas sukurti jos vidinį generuojamąjį AI. Vasario mėnesį bendrovė atšaukė dešimtmečius trukusias pastangas sukurti jos elektromobilį. Kai kurie darbuotojai buvo perskirstyti į šiuos generuojančio AI projektus.

 

 „Apple“ ieško potencialių išorinių partnerysčių, susijusių su pažangesniu AI. Giannandrea ir Federighi susitiko su Samu Altmanu, OpenAI vadovu, sakė su diskusijomis susipažinę žmonės.

 

 „Wall Street Journal“ savininkas „News Corp“ bendradarbiauja su „OpenAI“ turinio licencijavimo srityje." [1]

1. Apple Fell Behind in AI Arena, Catching Up Is Goal. Tilley, Aaron.  Wall Street Journal, Eastern edition; New York, N.Y.. 07 June 2024: A.1.

Apple Fell Behind in AI Arena, Catching Up Is Goal

"For those who saw them, the demonstrations inside Apple earlier this decade of a revamped Siri offered a showcase of the amazing capabilities a powerful AI voice assistant could have.

The famed assistant, one of the last projects Apple co-founder Steve Jobs worked on before his death, had been given a total overhaul. Capable of running on an iPhone and without an internet connection, the new Siri impressed people with its improved speed, conversational capabilities and the accuracy with which it understood user commands. Code-named Project Blackbird, the effort also imagined a Siri with capabilities built by third-party app developers, according to people familiar with the work.

Yet a competing project won out in an internal contest ahead of the 10-year anniversary of Siri's launch. Known as Siri X, the more-modest upgrade involved moving more existing Siri software onto iPhones from remote servers to improve the voice assistant's speed and privacy. The Siri X enhancement was unveiled in 2021.

Next week, at Apple's annual Worldwide Developers Conference, the company is set to join an AI arms race that many think will define the future of technology. The iPhone maker is trying to catch up with Microsoft, Alphabet's Google and other rivals that have begun to integrate generative AI into their core products.

Apple's caution and characteristic secrecy, as well as the care it takes in upgrading devices -- where hardware and software are seamlessly integrated -- have hobbled its early efforts in the AI arena, the people said. It now finds itself in the unusual position of having to take risks.

The company is set to announce an array of generative AI upgrades to its software products, including Siri, people familiar with its plans said. The AI features include assistance in message writing, photo editing and summarizing texts.

While Apple isn't expected to catch up with leading AI innovators soon, the company is preparing to unveil AI features with impressive capabilities that also maximize privacy for users -- a concern that will be central to unlocking the full potential of AI assistants. 

Apple is also expected to unveil one or more potential partnerships with major AI developers after holding talks with OpenAI, Google and Cohere, the people said.

Apple has long prided itself on perfection in its product rollouts, a near impossibility with emerging AI models. While OpenAI systems have dazzled more than 180 million users with their generation of writing, images and video, they are prone to occasional errors, often called hallucinations. Apple has had limited tolerance for such issues.

"There's no such thing as 100% accuracy with AI, that's the fundamental reality," said Pedro Domingos, a professor emeritus of computer science and engineering at the University of Washington. "Apple is not compatible with that. They won't release something until it's perfect."

Apple has weighed whether to allow users to choose a third-party AI provider that could supplant or help power Siri, one of the people said. It is unclear in what way Siri could be powered, augmented or replaced by third-party AI providers, or whether Apple will move forward with a version of that idea.

Bloomberg earlier reported on an OpenAI partnership, and the Information reported on efforts to overhaul Siri.

Google, Microsoft and Samsung Electronics have begun rapidly integrating generative AI into their devices and services. While Apple finds itself behind in a generational shift taking place in the tech industry, many investors and AI experts have said the company will find a way to bring generative AI to the masses.

"Apple can do pretty much anything they set their minds to," said Vineet Khosla, a former engineering manager on the Siri team who is currently chief technology officer at the Washington Post. "There is a consumer focus that exists at the company. The focus of their AI is to make it work in a very privacy-sensitive manner."

Over the years, Apple has made improvements to Siri and incorporated smaller AI features throughout all of its products. In the recently released headset Vision Pro, AI is extensively used for tracking eyes and hand positions.

When Siri launched in 2011, Apple was ahead of rivals in seeking to establish the first AI assistant. Jobs, who spearheaded the acquisition in 2010 that led to Siri, encouraged the team to keep the assistant's dry wit and sense of humor. The early launch demonstrated the company's willingness to take risks.

"Siri was the last thing Apple was first on," said Dag Kittlaus, co-founder of the Siri startup that Apple bought. He left Apple soon after the product's launch.

As Apple struggled to keep advancing Siri, the company hired one of Google's top engineering executives to run its AI strategy: John Giannandrea. In 2018, he was elevated to the role of senior vice president, reporting directly to Apple Chief Executive Tim Cook.

In early team meetings, Giannandrea made it clear that improving Siri was a main focus. He was also tasked with centralizing much of Apple's fragmented efforts regarding AI. He began building up his AI team by recruiting Google employees and through startup acquisitions, but his team had difficulty fitting in with the rest of Apple, according to people familiar with their work.

The new AI group operated much like parts of Google, where deadlines are more loosely defined. Teams inside Apple need to maintain rigorous deadlines to have products ready for release events every fall. Efforts to collaborate between other parts of Apple that were building products and the AI team at times fell apart because they couldn't agree on deadlines, the people said.

Instead of working with the AI team, other parts of Apple busy building software products maintained their own, separate AI capabilities. For example, the software group run by Senior Vice President Craig Federighi continued to invest in and build up the AI behind its image- and video-recognition capabilities, said former Apple employees.

When ChatGPT launched in late 2022, everything changed. Federighi, the software chief, became a convert that Christmas break after he began playing around with the Microsoft-owned GitHub AI coding tool called Copilot, which is powered by OpenAI's technology, said people familiar with his experience.

Apple stepped up efforts to build its own internal generative AI. In February, the company canceled its decadeslong effort to build its own electric car. Some of the employees had been redeployed into these generative AI projects.

Apple has been looking potential external partnerships for more advanced AI. Giannandrea and Federighi have met with Sam Altman, OpenAI's chief executive officer, people familiar with the discussions said.

Wall Street Journal owner News Corp has a content-licensing partnership with OpenAI." [1]

1. Apple Fell Behind in AI Arena, Catching Up Is Goal. Tilley, Aaron.  Wall Street Journal, Eastern edition; New York, N.Y.. 07 June 2024: A.1.

The biggest political problem in the eastern EU countries

 "GDP per capita here reached more than 70 percent from the EU average."

Prices in stores have become comparable to Western prices. But the average income (salaries and pensions) is several times lower than in the West [1]. Some people takes that money for themselves. Unions and politicians are needed to make a breakthrough so that the income in the eastern EU countries becomes lower than the income in the West by only less than 30 percent.

1. Michał Barc: "It’s true an average wage in Czechia is roughly half of that in Italy."

 
 
   

Didžiausia politinė problema rytinėse ES šalyse

"BVP vienam gyventojui čia pasiekė daugiau, nei 70 proc. nuo ES vidurkio."

Kainos parduotuvėse susilygino su Vakarų kainomis. Bet vidutinės pajamos (atlyginimai ir pensijos) yra keletą kartų  mažesnės, negu Vakaruose [1]. Kažkas pasiima tuos pinigus sau. Reikia profsąjungų ir politikų, pramušančių, kad pajamos rytinėse ES šalyse taptų mažesnės už Vakarų pajamas tik mažiau, nei 30 proc.

1. Michał Barc: "Tai yra tiesa, kad vidutinis atlyginimas Čekijoje yra maždaug perpus mažesnis, nei Italijoje."

 
 
  

Who needs qubits? Physicists make light-based ‘qumodes’ for quantum computing

"Careful retooling of laser beams allows scientists to harness photons for performing quantum calculations.

 The optical parametric oscillator used as the quantum light source.

 Quantum computers are built from qubits: physical objects with properties that have one of two possible results when measured. The prototypical example is an electron, whose spin will always show as either clockwise or counterclockwise.

 

Now physicists have demonstrated an alternative foundation for quantum computers called qumodes, which were first proposed in 2001. Their measured properties — in this case, the brightness of a light pulse — can vary along a continuum instead of consisting of just two discrete possibilities1.

 

To create qumodes, Shunya Konno at the University of Tokyo and his collaborators carefully modified laser pulses by removing one photon at a time and creating interference between pairs of pulses. They then demonstrated that the resulting pulses had the properties that would be required, both to perform ‘digital’ quantum computations and to correct errors in those computations.

 

This means that, with further work, a few photonic qumodes could perform the same kinds of quantum algorithm that require hundreds or even thousands of qubits in ordinary quantum computers.

 

Theoretical studies have suggested that qumode-based photonic quantum computers could be faster than qubit-based machines, and easier to build at large scales without becoming impractically prone to computational errors." [1]

1. Nature 626, 11 (2024)

Kam reikalingi kubitai? Fizikai kuria šviesos pagrindu sukurtus kvantinio skaičiavimo „kumodus“

 „Atsargus lazerio spindulių pertvarkymas leidžia mokslininkams panaudoti fotonus kvantiniams skaičiavimams atlikti.

 

 Optinis parametrinis osciliatorius, naudojamas, kaip kvantinis šviesos šaltinis.

 

 Kvantiniai kompiuteriai yra sukurti iš kubitų: fizinių objektų, kurių savybės turi vieną iš dviejų galimų rezultatų išmatuojant. Prototipinis pavyzdys yra elektronas, kurio sukimasis visada bus rodomas pagal laikrodžio rodyklę arba prieš laikrodžio rodyklę.

 Dabar fizikai pademonstravo alternatyvų kvantinių kompiuterių principą, vadinamų kumodais, pagrindą, kuris pirmą kartą buvo pasiūlytas 2001 m. Jų išmatuotos savybės – šiuo atveju šviesos impulso ryškumas – gali kisti išilgai kontinuumo, o ne sudaryti tik dvi atskiras galimybes.

 Norėdami sukurti kumodus, Shunya Konno iš Tokijo universiteto ir jo bendradarbiai kruopščiai modifikavo lazerio impulsus, pašalindami po vieną fotoną ir sukurdami interferenciją tarp impulsų porų. Tada jie įrodė, kad gauti impulsai turi savybių, kurių prireiks „skaitmeniniams“ kvantiniams skaičiavimams atlikti ir tų skaičiavimų klaidoms ištaisyti.

 Tai reiškia, kad, toliau dirbant, keli fotoniniai kumodai galėtų atlikti tokius pačius kvantinius algoritmus, kuriems įprastuose kvantiniuose kompiuteriuose reikia šimtų ar net tūkstančių kubitų.

 Teoriniai tyrimai parodė, kad kumodais pagrįsti fotoniniai kvantiniai kompiuteriai gali būti greitesni už, kubito pagrindu, veikiančias mašinas ir lengviau sukurti kvantinį kompiuterį, veikiantį dideliu mastu, netapusį nepraktiškai linkusiu į skaičiavimo klaidas." [1]

 

1. Nature 626, 11 (2024)