Inconsistency between questions: in Time From (weak) AGI to Superintelligence the community forecasts 2.5 years, which combined with the current forecasted date for Date Weakly General AI is Publicly Known (Dec 28, 2027) gives Superintelligence by 2031 (in contrast with 2038 predicted here for AGI).

Shouldn't this question be resolved? I think even GPT-3 got a decent chance of completing the task and since its debut there appeared even more capable models. If this doesn't qualify as "actual demonstration", it would be nice to get an assessment from "two experts in the field that a comparably capable system exists" (as mentioned in resolution criteria).

(Conflict of interest: I put 8% of the cumulative probability that this will happen before 2022.)

[New Law Gives Sweeping Powers To Hungary's Orban, Alarming Rights Advocates](https://www.npr.org/sections/coronavirus-live-updates/2020/03/30/823778208/new-law-gives-sweeping-powers-to-hungarys-orban-alarming-rights-advocates?t=1585603508076) >The nationalist government in Hungary passed a law Monday granting sweeping emergency powers that Prime Minister Viktor Orban says are necessary to fight the coronavirus pandemic. >Those powers include sidelining parliament and giving Orban the power to rule by decree indefinitely. The law would punish those who...

@JurijZucker wrote:

As I said last year, there is 0% that SpaceX will land people on Mars IMO.

Sigh... Would you be willing to bet your $10 000 against my $10?

From one of the authors of Cicero:

It's designed to never intentionally backstab - all its messages correspond to actions it currently plans to take. However, sometimes it changes its mind...

Seems like a limitation that when lifted could boost performance.

@Skyt3ch Would you take a bet where you get $10 if Trump is not re-elected and pay out $200 if he is re-elected?

[**Human-Timescale Adaptation in an Open-Ended Task Space**](https://sites.google.com/view/adaptive-agent/) by DeepMind >Foundation models have shown impressive adaptation and scalability in supervised and self-supervised learning problems, but so far these successes have not fully translated to reinforcement learning (RL). In this work, we demonstrate that **training an RL agent at scale** leads to a general in-context learning algorithm that **can adapt to open-ended novel embodied 3D problems as quickly as humans**. >In a vast space of held-out env...

@Matthew_Barnett Does this question consider only dense models or do sparse models also qualify for question resolution?

[ETA: **This approach (using calculators) does not satisfy the resolution criteria.**] ["A neural network solves, explains, and generates university math problems by program synthesis and few-shot learning at human level"](https://www.pnas.org/doi/10.1073/pnas.2123433119) Abstract: >**We demonstrate that a neural network pretrained on text and fine-tuned on code solves mathematics course problems, explains solutions, and generates questions at a human level. We automatically synthesize programs using few-shot learning and OpenAI’s Codex transformer and...

Sorry for venting but assigning rapidly rising probability per launch opportunity between 2027 and 2040 is a nightmare when using sliders and available years go up to 2100. Can we get this question as a set of binary questions per Hohmann transfer window or as a range question with years going up to 2050?

[**Human-Timescale Adaptation in an Open-Ended Task Space**](https://sites.google.com/view/adaptive-agent/) by DeepMind >Foundation models have shown impressive adaptation and scalability in supervised and self-supervised learning problems, but so far these successes have not fully translated to reinforcement learning (RL). In this work, we demonstrate that **training an RL agent at scale** leads to a general in-context learning algorithm that **can adapt to open-ended novel embodied 3D problems as quickly as humans**. >In a vast space of held-out env...

@(isinlor) wrote: > @(Spirit59) I'm the author of the corresponding NASA question, but not the author of this question. > > I have the same expectation as you for this question. To resolve positively SpaceX logo MUST be prominent on the mission hardware. Obviously, this is what "branded" means. But more implicitly that it also should be in big part SpaceX effort, at least SpaceX rocket and spacecraft. > > I would not add to this question any financial requirements. > > I think community took similar interpretation. At least people who tried to actually...

In a [similar question concerning NASA](https://www.metaculus.com/questions/1476/will-nasa-land-people-on-mars-prior-to-2030/) word "branded" seems to be synonymous with a particular level of financial contribution: >The question will resolve positively even if the NASA-branded mission makes use of SpaceX transport system, under the condition that the main funding for the mission comes from USA budget. also, from the comment section: >>what if a multinational effort with NASA a major but not over 50% contributor? >If not over 50%, then NASA would nee...

@(randomuser2323)This comment gives the impression that the planned uncrewed mission in 2018 and crewed in 2024 was meant to be performed with the same vehicle which it wasn't - it's important to note that 2018 mission was planned with [Red Dragon](https://en.wikipedia.org/wiki/SpaceX_Red_Dragon) in mind, not Starship. >In April 2016, SpaceX announced that they had signed an unfunded Space Act Agreement with NASA, providing technical support, for a launch no earlier than 2018. In February 2017, SpaceX noted this launch date was delayed to no earlier than...

[**RT-1: Robotics Transformer for Real-World Control at Scale**](https://ai.googleblog.com/2022/12/rt-1-robotics-transformer-for-real.html) >[...] we propose the Robotics Transformer 1 (RT-1), a multi-task model that tokenizes robot inputs and outputs actions (e.g., camera images, task instructions, and motor commands) to enable efficient inference at runtime, which makes real-time control feasible. This model is trained on a large-scale, real-world robotics dataset of 130k episodes that cover 700+ tasks, collected using a fleet of 13 robots from Everyda...

https://twitter.com/DanHendrycks/status/1395536919774121984 >Can Transformers crack the coding interview? We collected 10,000 programming problems to find out. GPT-3 isn't very good, but new models like GPT-Neo are starting to be able to solve introductory coding challenges. From the paper's abstract: >[...] Our benchmark includes 10,000problems, which range from having simple one-line solutions to being substantial algorithmic challenges. We fine-tune large language models on both GitHub and our training set, and we find that the prevalence of syntax e...

1. OpenAI Model Generates Python Code

2. GPT-3 JavaScript tutorial taken from here

I think this question should be added to Elon Musk related timeline. [edit: also these questions [1] [2] [3]

— edited by Spirit59

What happened to Forecasting AI Progress tournament? It's not present on the list of tournaments. Its original page is blank. On the page with tournament rules timeline is deleted (IIRC what I read from archived page on waybackmachine, results should be announced mid-February this year). I asked about it on Twitter by tagging metaculus Twitter profile but got no response. What is going on?

@jacob.steinhardt This question should get resolved for "June 30, 2023".