berkeley-nest

/

Starling-LM-7B-alpha

like 553

Follow

Berkeley-Nest 61

Thank you! Yes we also observe that in rare cases the model just doesn't know when to stop. We're selecting a better checkpoint as beta version right now.

May I ask if you can share your prompt that leads to such behavior? That will help us debugging the new model as well. Thanks!

Sure. Here is the prompt. I tested this in lmsys chat. Max tokens = 512. I checked for llama2 tokens = 2323 tokens

GPT4 Correct User: Strategy
How Tesla’s Charging Stations Left Other Manufacturers in the Dust
by Hemant Bhargava, Jonas Boehm, and Geoffrey G. Parker January 27, 2021
VCG/ Getty Images
Summary. Mainstream car manufacturers are spending billions on developing EVs. Before spending any more money they should take a leaf out of Tesla’s books and create a recharging network to support their cars. Only then will they be able to treat their EVS as a product.... more

Over the past five years, the major auto companies have invested massively in electric vehicles (EVs). The Volkswagen Group in 2017 announced that they would offer 80 new electric vehicles across their brands by 2025 and electric versions of every one of its models by 2030. In the same year GM went public with plans to put at least 20 new electric models on the road by 2023. They are not alone; Bloomberg New Energy Finance predicts that 500 different EV models will be available globally by 2022.
Yet, despite investments that add up to many billions of dollars, none of the major incumbent automakers seems to pose much of a threat to market leader Tesla, which has become nearly synonymous with EVs. This is surprising since one might reasonably have expected that once firms with annual revenues in excess of $100B, deep manufacturing expertise, and large market shares turned their attention to the electric vehicle market, the game would be up.
The reason why consumers still choose Teslas over products like Audi’s eTron or attractive EVs from GM’s Buick, Cadillac, GMC, and Chevy brands is perhaps surprisingly simple. They can drive their Teslas for long distances in full confidence that they will find convenient locations at which to recharge their vehicle. While the incumbent automakers are still focused narrowly on perfecting their electric cars, Tesla has been thinking about the entire vehicle system, with the aim of solving consumers’ core driving needs.
The car as platform
A car creates value to its owner when it is driven, which requires refueling. Automakers of gasoline cars or trucks do not have to worry about this, as refueling stations are abundant — over 160,000 stations are in the United States alone — and easily accessible. They have therefore built their strategies around standard marketing variables: product, price, placement, promotion. Build a great car (or truck), advertise it heavily, offer it in the right markets at a good price, and the product will sell.

An electric car, however, requires a different value analysis. “Refueling stations” — i.e., rapid charging facilities — for electric vehicles are in their infancy, with only about 4,000 available in the United States. Moreover, the network of available charging stations is highly fractured across ownership and technology. The next largest network, compared to Tesla, is only 10% as large. Unless you buy a Tesla, you have few options for reliable route planning, guaranteed access, and rapid public charging.
An electric car, therefore, is a two-sided platform good, the two sides being an installed base of car buyers and a large network of geographically dispersed multi-stall rapid-charging stations. Selling electric cars requires a robust charging network. But, investments in building a massive charging network make sense only if there is a large enough user base and demand for these chargers. Tesla has such a network, and everyone else’s is laughable. How did this happen and what can Tesla’s history teach us?
Platforms need networks
Nissan, with the zippy and relatively affordable Leaf, stole an early lead in the EV market, and was the best-selling electric car from 2011 to 2014. Despite this lead, Nissan failed to provide a robust fast- charging network which left buyers relying on a small number of third-party stations available to all brands.
Tesla’s approach was strikingly different. They began with a vanity product, the Roadster, that got them off the ground and generated some early sales. They then moved on to release the Model S in 2012, which for its few years (2013-15) had roughly a one-year waiting period. To support the cars, however, Tesla rolled out a coast-to-coast proprietary network. So, although Tesla sold only a few thousand cars in its early years, it had built out a huge network. This addressed the buyer’s “range anxiety” problem — no one who considered buying a Tesla needed to worry much about charging.

Most automakers have followed Nissan’s approach and are focusing investments on making better electric cars. But imagine if, instead of investing tens of billions of dollars in producing cars with no way to drive long distances individually, Audi, GM, Ford, and the rest each spent just a billion dollars to build a network of supercharging stations. In North America, that amount would finance approximately 1,000 locations with 10 charging stalls each. If the stations were correctly placed, a network of that size should give buyers enough confidence to choose a car based on its features instead of on the features and the charging network. Then firms could begin the work of reaching viable volumes, bringing costs down, and eventually becoming serious competitors to Tesla.
The platform advantage
Employing a proprietary platform strategy, as Tesla did, enables the platform owner to coordinate the two sides of the market: the installed base of cars and the network of charging stations. Because it owns the charging network, Tesla can choose how to price (whether to make charging free and monetize only the car), the number of stations, rollout timing, and location.
These choices can reflect Tesla’s overall business strategy and detailed knowledge of where the buyers are and where they drive. Interestingly, another newcomer, Rivian (which has yet to sell a single vehicle) is also building a proprietary charging network, like Tesla’s. Rivian is splitting its stations between major highways and campgrounds, a perfect fit given its focus on electric adventure vehicles.
Automakers would be well advised to take a leaf out of Tesla’s playbook and focus on the network before heavily investing in designing and manufacturing new EVs, or at least do so in parallel. They may not actually have to build the network — they could instead partner with firms that have networks that could accommodate

recharging stations. Many existing fossil fuel energy firms, for example, have gas station assets that will eventually become stranded and could be repurposed for electric vehicles.
Focusing on the network, of course, is not without its risks. Rolling out a network from scratch isn’t a trivial challenge, nor is it clear that potential partners would be willing to commit to an exclusive relationship with any one automaker, which the latter would need, at least to begin with, in order to get a head start on the other automakers. But investing in a network will certainly increase the odds of winning a dominant position in EVs, which a focus on the
cars alone is unlikely to deliver — at least according to the evidence so far.
Looking ahead
It’s clear that Tesla itself is doubling down on its Big Tech platform strategy. At present, its business model for its new automated driving capability is classic product pricing; levy a one-time upcharge of $10,000. However, it plans to switch to selling automated driving as a service for a monthly fee. This strategy implicitly defines the car itself as a platform on top of which services can be delivered.
Such a business model offers the additional benefit of enabling Tesla to gather the training data that the machine learning algorithms for true self-driving cars will need, which will confer a critical advantage in the next stage of motor vehicle competition. As for the competition around charging networks, should other firms get serious about building out alternatives, then we would expect Tesla to open its own network since the advantage to remaining closed would start to fade. Indeed, we are beginning to see initial signs of openness as Tesla hints at allowing a new partner to connect.
Firms that dream of being the next Tesla should carefully examine why they are so far behind. It’s not for lack of knowledge about how to build cars; many of the incumbent firms have been doing that for over a hundred years. Instead, they should focus on the critical infrastructure, in this case the charging networks, that have made

customers willing to take a chance on a newcomer. And, once they’ve done that, they can begin to address the next battleground: control over the vehicle data that will enable self-driving cars and the transition to vehicle as a service instead of vehicle as a product.
HB
Hemant K. Bhargava is the Jerome and Elsie Suran Chair in Technology Management at the University of California Davis Graduate School of Management and Director of the Center for Analytics and Technology in Society
JB
Jonas Boehm is Venture Strategist & Platforms Lead at AGL Energy and Fellow at the World Economic Forum’s Global Future Council on Advanced Manufacturing and Production
Geoffrey G. Parker ([email protected]) is a professor of engineering at Dartmouth College
and a research fellow at MIT’s Initiative on the Digital Economy. He co-authored Platform Revolution (W.W. Norton & Company, 2016), the April 2016 HBR article “Pipelines, Platforms, and the New Rules of Strategy” and the October 2006 HBR article “Strategies for Two-Sided Markets,”
an HBR all time top 50. Follow him on Twitter @g2parker.

Can you summarize the above article?<|end_of_turn|>
GPT4 Correct Assistant:

Tesla's success in the electric vehicle (EV) market is largely due to its focus on creating a robust charging network to support its cars. While other mainstream automakers have been investing heavily in developing EVs, they have not focused as much on the charging infrastructure, resulting in a less attractive proposition for consumers. By building a comprehensive charging network, Tesla has addressed the "range anxiety" problem that has plagued the EV market. Major automakers should take a leaf out of Tesla's book and focus on creating a reliable charging network before heavily investing in designing and manufacturing new EVs. This will increase their chances of winning a dominant position in the EV market, as a focus on the cars alone is unlikely to deliver the same results.

In the battle of the electric vehicle (EV) charging network, Tesla has taken a massive lead over traditional automakers, and this advantage is unlikely to fade anytime soon. While other carmakers have been focused on perfecting their electric cars, Tesla has been thinking about the entire vehicle system. It's not just about the cars; it's about the charging infrastructure that supports them.

Tesla's approach was to start with a vanity product, the Roadster, which generated initial sales and allowed them to develop a substantial charging network. Although Tesla sold only a few thousand cars in their early years, they had built out a large charging network, addressing the "range anxiety" problem and giving buyers the confidence to choose a Tesla based on its features rather than the charging network.

On the other hand, most automakers have focused their investments on making better electric cars without paying much attention to the charging infrastructure. This has left them trailing far behind Tesla in terms of charging network strength. It's worth noting that Tesla's charging network is proprietary, which gives them control over the pricing, number of stations, rollout timing, and location. This strategy is a significant advantage over the fragmented and fractured charging networks of other automakers.

To catch up with Tesla, traditional automakers need to shift their focus from just the cars to the charging network. They could either build their charging networks from scratch, which is a challenging task, or they could partner with existing firms that have the infrastructure in place. Some potential partners could be fossil fuel energy firms that have gas station assets that will eventually become obsolete and could be repurposed for electric

Thank you! Actually this one looks fine for me on lmsys. See my screenshot below.

In lmsys chat, did you also include the chat template like "GPT4 Correct Assistant"? This won't be needed there since they're automatically set up by lmsys.

I tried several generation with temperature = 0.7, which seems fine. But it's likely that with a high temperature the model will generate a bit longer. You can try temperature = 0 to see if things still work like this.

I tried again with your recommendation. Removing the prompt template and feeding as one text block. And here is the below params and generation. The summary is really good but just eos is not generated sometimes

Mainstream car manufacturers have invested heavily in electric vehicles (EVs), but have not yet posed a significant threat to market leader Tesla. Tesla's success lies in its focus on the entire vehicle system, solving consumers' core driving needs by creating a robust charging network. Automakers have focused on improving their electric cars, but have not invested in building a charging network. Tesla's approach has given it a significant advantage, and other automakers should focus on building charging networks in order to compete effectively in the EV market.

In the past five years, major auto companies have invested billions in developing electric vehicles. Despite these investments, none of the major incumbent automakers have posed much of a threat to market leader Tesla. This is surprising, as one might reasonably have expected that once firms with annual revenues in excess of $100 billion, deep manufacturing expertise, and large market shares turned their attention to the electric vehicle market, the game would be up.

The reason why consumers still choose Teslas over products like Audi's eTron or attractive EVs from GM's Buick, Cadillac, GMC, and Chevy brands is perhaps surprisingly simple. They can drive their Teslas for long distances in full confidence that they will find convenient locations at which to recharge their vehicle. While the incumbent automakers are still focused narrowly on perfecting their electric cars, Tesla has been thinking about the entire vehicle system, with the aim of solving consumers' core driving needs.

A car creates value to its owner when it is driven, which requires refueling. Automakers of gasoline cars or trucks do not have to worry about this, as refueling stations are abundant — over 160,000 stations are in the United States alone — and easily accessible. They have therefore built their strategies around standard marketing variables: product, price, placement, promotion. Build a great car (or truck), advertise it heavily, offer it in the right markets at a good price, and the product will sell.

An electric car, however, requires a different value analysis. “Refueling stations” — i.e., rapid charging facilities — for electric

Oh I see. It's interesting that when I include the "Strategy" at the beginning, the generation seems to be shorter. And when I remove that I can reproduce what you had.

It looks like the model is trying to be extremely verbose here and doesn't understand how to stop. Thank you for the prompt! I also did notice that if you say "Please summarize the article in 100 words", its output seems to be shorter lol. We will try to fix this in the beta version.