1) Ok so dramatic new revelations (not) — After OpenAI became ClosedAI, and sold out to Big Tech, it stopped being so open after all … egads, my faith in the benevolence and transparency of Silicon Valley elites is shattered ;p technologyreview.com/s/615181/ai-op… @_KarenHao
2) Those with longer than average memories may recall that a few years ago, OpenAI was funded by Musk and Sam Altman with a narrative of guiding AGI development in an open and beneficial direction… but from the start they were clear about their non-commitment to open source
3) OpenAI was founded with a $100M dedication of funding from Musk and Altman and others, but it was always a bit waffly —i.e. it wasn’t $100M put into an OpenAI account, it was some sort of MOU roughly committing conditionally for the future based on needs and progress
4) OpenAI started off following more Tensorflow-style than Linux-style OSS, i.e. closed strategy and design leading to periodic polished OSS releases made by a closed team
5) OpenAI articulated a core methodology of taking AI ideas from the prior literature, and trying to make them work better via throwing huge masses of money and compute resources at them
6) OpenAI made some genuine cool, exciting advances in applications of deep NNs — though nothing fundamentally innovative on the order of, say, the introduction of Transformer networks or CNNs
7) When OpenAI launched the GPT2 language model — a very nice incremental advance in transformer neural nets for NLP, in the series of many other similar advances by others —
they made a lot of bogus media noise about how they were keeping it secret due to its potential risks.
they made a lot of bogus media noise about how they were keeping it secret due to its potential risks.
8) The last few naive fools believing OpenAI was in any meaningful sense open, shed a few bitter tears
9) Shortly after the bogus media noise about GPT2’s dangers, OpenAI secures $1B from Microsoftt, to be disbursed according to an unclear and non-disclosed schedule, as part of an arrangement including OpenAI using Azure for its computing
10) Lo and behold, it is revealed that this elitist, secretive, brilliant, well-resourced, exciting, big-tech-company-controlled AI project that misleadingly has “open” in its name, is not so open after all…
11) Meanwhile, those of us pushing for democratic, decentralized, participatory, ACTUALLY open approaches to AGI … are working hard and having some deep successes, but not yet making as many headline-making demos because of our vastly smaller hardware budgets
12) The tendency toward secrecy & corporate oligopoly that OpenAI illustrates, is a deep issue that ties in with problems in the global economy regarding wealth and power and information inequality — and these issues will get more and more severe as Narrow AI verges into AGI
13) How to do it better: Open processes as well as open source products, blockchain based decentralized networks, investment and biz model strategies not focused on exit to Big Tech, multi-paradigm integrative AI rather than deep-NN narrow-focus @singularity_net #opencog
14) What an uphill climb it is for those of us pushing toward genuinely beneficial, open, democratic, compassionate AGI — we are competing with not just military and surveillance AI efforts, but also corporates using the verbiage of open-ness and benefit for their own ends
15) To be clear: military and intel and huge corporations all will have advanced AI. But they shouldn’t be dominant — we need the center of gravity of AGI two be in open, benefit-oriented, participatory networks. This is how to work toward a positive Singularity.
3*) Ah whoops I noticed a typo above: actually OpenAI was founded in 2015 with a dodgily phrased "commitment" of $1B, not just $100M. But I mean hey, what's $900M more or less when you're Masters of the Universe right? ;D venturebeat.com/2015/12/11/sam…
typo:$1B not $100m was the initial dedication, woops
