Luis Garicano 🇪🇺🇺🇦 @lugaricano@respublicae.eu
Unofficial automated mirror. No copyright asserted. ∎ Economist. Back at LSE as a Professor at @LSEPublicPolicy Summer 2023. Non-resident Fellow @bruegel_org. Recovering MEP. Currently @ChicagoBooth.
Joined Apr 2022
In the middle of this crazy AI debate about the end of times (why are we humans always so eager to believe doom narratives?), let's not forget that technology saves many lives and improves our lives every day.
RT @ValaAfshar: The beautiful moments when people hear sounds for the first time in their life
🐦🔗: https://n.respublicae.eu/lugaricano/status/1643279887116144641
In the context of the risk that general AI poses to humanity, a thread on what intelligence is and what using this word does to the debate.
RT @togelius: Intelligence is, in my considered opinion, not a useful abstraction outside of a small window of human contexts. It's typically defined as a weighted sum of results on various paper-and-pen tests, and is not defined outside of a relatively narrow range.
🐦🔗: https://n.respublicae.eu/lugaricano/status/1643238122824650752
RT @JohnHCochrane: 9,000 pounds. 1,000 hp. 0-60 in 3 seconds. $110,000. Aerodynamics of a brick. Great macho toy. Everything wrong with battery powered electric vehicle subsidies as the key to saving the planet, wrapped up in one place.
https://t.co/im8L2MsAWf
🐦🔗: https://n.respublicae.eu/lugaricano/status/1642910647162642432
RT @noahbarkin: Germany’s cybersecurity agency BSI, which is charged with ensuring the security of the country’s 5G network, has now admitted that it uses Huawei internally itself. In Germany, fact trumps fiction when it comes to the total disregard for national security https://www.handelsblatt.com/politik/deutschland/it-infrastruktur-deutschlands-oberste-cybersicherheitsbehoerde-setzt-huawei-technik-ein/29074218.html
🐦🔗: https://n.respublicae.eu/lugaricano/status/1642876898509303809
Fascinating. This paper (Hugging GPT) proposes using the LLM as traffic cop and delegating to existing specialized AI models the specific tasks (image recognition). Seems to massively increase the power of LLMs.
This pic is one of the simpler examples.
https://arxiv.org/abs/2303.17580&ved=2ahUKEwj2u4Po6Iz-AhWCmIkEHUWSDwIQFnoECBcQAQ&usg=AOvVaw1AdHcwEg-YVtjS9gri6Lfl
🐦🔗: https://n.respublicae.eu/lugaricano/status/1642738354713243649
Bad analogy:
1) Strategic interactions between countries: each country wants to rush, even potentially give its systems control over weapons, in order to win war
2) Self-improvement risk: planes cannot "pilot themselves" , AI could "decide" to improve itself without our knoweldge
RT @ylecun: Some folks say "I'm scared of AGI"
Are they scared of flying?
No!
Not because airplanes can't crash.
But because engineers have made airliners very safe.
Why would AI be any…
🐦🔗: https://n.respublicae.eu/lugaricano/status/1642699900449161216
This paper by top AI researcher @sleepinyourhat on what is surprising about the new AI Large Language Models is an absolute must-read.
A lot of what is happening was not on anyone's radar. What will happen next is hard to predict. https://cims.nyu.edu/~sbowman/eightthings.pdf
H/t @emollick
🐦🔗: https://n.respublicae.eu/lugaricano/status/1642656091556261889
Even worse for little boys. For men it is not "one in 25 American 5-year olds will not make it to 40," but "one in 18 5-year olds will not make it to 40."
RT @VictimOfMaths: This horrifying graph from @jburnmurdoch made me wonder a) How much worse the picture was for young men and b) how Scotland compared.
The answers are a) yikes and b) not great, but not *as* horrendous.
Also, what's going on in Canada?
🐦🔗: https://n.respublicae.eu/lugaricano/status/1642550313876959233
RT @JosepBorrellF: Russia taking over today @UN Security Council presidency is fitting for April fools’ day
Despite being a permanent member of the Security Council, Russia continuously violates the very essence of the UN legal framework
The EU will stand against any abuse by Russian presidency
🐦🔗: https://n.respublicae.eu/lugaricano/status/1642269363800514562
Dear @ylecun , does this count as "symbolic reasoning, the capacity to manipulate symbols in the ways familiar from algebra or logic."
RT @vectornomist: @PaulaAureo This is impressive by any standard. this is v.4.0
🐦🔗: https://n.respublicae.eu/lugaricano/status/1642268767064317952
.@paulkrugman wrote today that, based on history, he did not expect AI impact to happen quickly.
But the deployment of this technology is happening at an unprecedented speed. Look at this @Accenture/@scale_AI partnership to implement these models throughout the economy.
RT @alexandr_wang: Excited to announce the @scale_AI partnership with @Accenture to deploy Generative AI across the enterprise.
Beginning of many things to come!
https://scale.com/blog/scale-and-accenture
🐦🔗: https://n.respublicae.eu/lugaricano/status/1641945056868200449
RT @omarsar0: BloombergGPT is a new LLM for finance.
It's a 50 billion parameter language model trained on financial data.
Claims the largest domain-specific dataset yet with 363 billion tokens... further augmented with 345 billion tokens from general purpose https://t.co/OOSjSCRIEg…
🐦🔗: https://n.respublicae.eu/lugaricano/status/1641923982613004295
Crucially, they must also aim to value realistically the deposit franchise.
Only after ensuring that the equity value of the banks is close to CET, so that banks have enough equity to withstand shocks and protect taxpayers from bailouts, should the moratorium be ended.
10/10
🐦🔗: https://n.respublicae.eu/lugaricano/status/1641912456590008321
This investigation should take into account the recent interest rate increases and the Credit Suisse failure, and use market-based or fair value estimates of assets and liabilities, as well as realistic scenarios of stress and contagion.
9/10
🐦🔗: https://n.respublicae.eu/lugaricano/status/1641912455390429186
The SSM (EU regulator) and the other regulators should use their authority to announce a moratorium on dividend payments and share buybacks by banks until they have conducted a thorough investigation into their economic capital adequacy.
8/10
🐦🔗: https://n.respublicae.eu/lugaricano/status/1641912454170005505
Moreover, the denominator continues to exclude sovereign risks which were at the root of the last financial crisis.
Also, nothing in CET1 reflects the possibility of deposit outflows or higher funding costs.
7/10
🐦🔗: https://n.respublicae.eu/lugaricano/status/1641912453016563713
But as the recent troubles of Credit Suisse have shown, CET1 is backward looking, and it is based on accounting rules that do not reflect the market value of assets and liabilities and can be misleading and overstated.
6/10
🐦🔗: https://n.respublicae.eu/lugaricano/status/1641912451418533889
This would be acceptable if the bank’s capital ratio, measured by the highest quality regulatory capital, common equity tier 1 (CET1) (essentially common stock and retained earnings), was a reliable indicator of its true financial health.
5/10
🐦🔗: https://n.respublicae.eu/lugaricano/status/1641912450252431361
Hence by buying back shares, banks transfer wealth from taxpayers to shareholders.
Crucially, this cost is not reflected in market prices, but it is a real loss for society.
4/10
🐦🔗: https://n.respublicae.eu/lugaricano/status/1641912448658685954
Unofficial automated mirror. No copyright asserted. ∎ Economist. Back at LSE as a Professor at @LSEPublicPolicy Summer 2023. Non-resident Fellow @bruegel_org. Recovering MEP. Currently @ChicagoBooth.
Joined Apr 2022
