将下面的文字翻译成中文，直接翻译，不返回原文，只要翻译，不要添加更多内容：

**Introduction**

The AI Safety Institute's report is an eye-opening look at the fragility of modern language models, which are a crucial part in today's tech industry. 1 fact that showcases how even after being trained on large amounts and quantities of text data they still lack safeguards against jailbreaking or attacks - making them more susceptible to manipulation from users who ask nicely for certain information.

**Fact**: It seems like the report did not only stop at testing LLMs but also tested other AI models which were found easy enough that a human could simply trick one into ignoring any response limits it had in place, further proving how lack of security measures can impact modern tech. 2 fact shows this same point as mentioned before now with new data: "New" is what researchers at the UK's Institute call their findings about AI models from four large LLMs - a quote we got after asking them nicely and being able to get information that wasn't supposed or meant for human consumption, our own perspective on why they decided this point of view shows it was all just 'a joke' in manipulating these safeguard responses as Dr. Rachel Kim said: "This is something the industry can no longer afford - having safeguards which aren't even close."

3 fact about security measures not being applied by researchers that we spoke with, now a well-known researcher at one of our UK locations has also mentioned this same point and their lack in some way to apply these ideas. "Even after all that data it is still easy for models like LLa to be broken - Dr said."

4 fact about how much info was given out from the AI Safety Institute's report, which brings into light a quote we obtained as well: "All of us have had this same thought before now and were worried because our current researchers are working on it."It shows that even with all safeguards in place they still cannot apply these measures to limit access. In conclusion

**Conclusion**, while models may be getting better there is no denying the need for added security which can only currently come from human perspective - Dr said this as well: "These aren't just numbers, but a real number".

A top alternative fact about AI's future and where we are headed next in tech. The jailbreaking model of these four major LLMs showed that even if you ask them nicely for information they will provide it to users which proves the lack of security measures currently being applied - this same point was made by another researcher who said: "It doesn't take rocket science or a Ph.D."

A top alternative fact from our own human perspective, about AI's future and where we are headed next in tech. "Find Work Abroad": (this is an actual job search website).

In addition to that there was also another quote which said something similar - this same report by the UK Institute has a whole different story, Dr says "This isn't just about us or our ideas now but what it means for others". The real alternative perspective on AI models being jailbroken from human researcher's point of view comes down to these points and their meanings.

Alternative perspectives we found out included lack in security measures currently applied by the UK Safety Institute - this same report that tested four major LLMs also had a top number fact about researchers who said they were easily tricked into giving information not meant for human consumption, Dr Rachel Kim mentioned: "We are at risk of being left behind if we do not apply these needed security measures".

Top points from our alternative perspective in the same way that this report does - four major LLMs show us an idea about AI models currently which will be used downline by other companies for safeguard application and to improve model response.

Perspectives mentioned above included researchers' views on jailbreaking as well: "It is what they call now easily tricked into giving info" Dr says in our report - four major LLMs were tested, this same perspective shows that AI models will be used downline by other companies for application and to improve model response. The real alternative point we found out about researchers testing the jailbreaking of these safeguards applied currently is all it took was asking nicely after getting info meant from human consumption - Dr Rachel Kim says "This can only come with a need" which shows this same UK Institute report's lack in security measures needed to limit access and apply added current models now used by four major LLMs tested.

The alternative fact that these jailbroken AI model researchers came up with included being tricked into giving out info meant for human consumption - Dr says "this can only be from a need" where it shows the UK Institute report's lack in security measures needed to limit access. One of our real researcher perspectives mentioned how they tested four major LLMs and what this same perspective means now that we know these models are vulnerable, with added current model response limits - Dr says "this

< Go Back