#Srivastava Group | Explore Tumblr Posts and Blogs

remesearch · 6 months

Text

Serial Fraudster: Gaurav Srivastava has been accused, and convicted, of fraud and misrepresentation multiple times in the US.

Convicted to pay USD 80,000 in 2019 and nearly USD 1 million in 2014, over the past few years, Gaurav has repeatedly demonstrated that he is a con artist.

Gaurav Srivastava, who has built a reputation of an honest and philanthropic businessman related to the CIA, has been revealed as a fraudster by online US legal data. Between 2019 and 2014, Gaurav and several of his family members have faced accusations and convictions multiple times in the US for fraud and misrepresentation, exposing highly questionable intentions behind their business dealings.

In 2019, Gaurav Srivastava was convicted by a Court order to pay USD 80,000 after failing to fulfil a payment obligation of USD 82,088 to the Keck Hospital in Los Angeles for medical services provided to his father, Jaswant Srivastava. Despite having issued checks from his company Veecon Biotech LLC (Delaware, US), Gaurav had later halted the payments. (Case No. 19STCV03981: https://www.lacourt.org/casesummary/ui/casesummary.aspx?casetype=civil)

Below is an excerpt from the legal documents of the Superior Court of California, Los Angeles County.

‘Gaurav Srivastava as the President of the Defendant company provided Plaintiff's Assignor with a check dated 05/24/18 in the amount of $14,850.00 made payable to Keck Medical Center USC and another check dated 05/24/18 in the amount of $67,238.00 made payable to Keck Medical Center USC for a total of $82,088.00 to pay for his father's medical services. Both checks were to be drawn from Defender company. The Defendant's President stopped payment on the checks and the balance remains unpaid’.

Gaurav Srivastava and his brother Pankaj Srivastava were also convicted by a Court order to pay nearly USD 1 million in 2014, when Nixon Peabody LLP (US) initiated legal proceedings against them and their company Veecon Group. They had breached an agreement made in 2013 by failing to honour invoices totalling USD 155,329 for legal services provided by Nixon Peabody LLP. As representatives of Veecon Group, Gaurav and Pankaj Srivastava were held accountable for the outstanding debt to Nixon Peabody LLP and were ordered to pay. (Case No. BC548034: https://www.lacourt.org/casesummary/ui/casesummary.aspx?casetype=civil)

Below is an excerpt from the legal documents of the Superior Court of California, Los Angeles County.

‘In early October 2013, in an effort to obtain Nixon Peabody's forbearance and willingness to continue as counsel, Gaurav Srivastava provided Nixon Peabody a personal check dated October 4, 2013 in the amount of $50,000.00. The check was returned for lack of sufficient funds. Later in October 2013, Gaurav Srivastava promised full payment by October 21, 2013. To date, no such payment, or for that matter, any payment has been received by Nixon Peabody’.

In 2017, Gaurav Srivastava was accused of fraud and breach of contract. Between 2015 and 2016, Veecon Biotech, under Gaurav Srivastava's leadership, entered into agreements with the plaintiffs, falsely claiming they held licenses to market saliva diagnostics technologies and appointing them as sales representatives. Gaurav Srivastava went further, making deceptive statements regarding Dr. David Wong's association with Veecon Biotech. If the plaintiffs withdrew their complaints after reaching settlements in October 2020, the Superior Court of California rejected these settlements, leaving the case unresolved. (Case no. BC648883: https://www.lacourt.org/casesummary/ui/casesummary.aspx?casetype=civil)

In 2019, a certain Patricia Riordan Torrey also initiated a legal case against Gaurav Srivastava, his company Veecon Biotech LLC, and his wife Sharon Srivastava as defendants. However, no additional information regarding the details or progression of the case could be located. Whilst Gaurav Srivastava and his wife Sharon Srivastava accused a certain Jill Small of defamation and of having 'intentionally interfered with a business deal that would have netted Plaintiff (Gaurav and Sharon Srivastava) approximately $40,000,000 in profit, and then tried to extort $500,000', the motion to seal the entire court file was denied in December 2022 because of a lack of specificity. (Case no. 21STCV28536: https://trellis.law/ruling/21stcv28536/gaurav-srivastava-vs-jill-small/20221208729619)

#fraud #gaurav srivastava #gaurav #gaurav kumar srivastava #con artist #fraudster #veecon group #sharon srivastava #srivastava #pankaj srivastava #scam

4 notes · View notes

cricketnationrise · 5 months

Note

Congratulations on 500!!! 🎉🎉 Extremely well deserved!

How about Zahra, 9:37pm, somewhere in England (I leave where exactly entirely up you)?

I'm on ao3 as hwaelweg 💙

for you my lovely, I have a lil date night moment for Zahra and Shaan. hope you love it, and thanks for being AMAZING and screaming about books with me 💜🦗

want your own ficlet? my followers can submit prompts based on these guidelines through January 31, 2024

❤️🤍💙❤️🤍💙

9:37pm, london

Zahra looks around from her perch at the bar. The pub is all dark wood and the smell of decades of smoke sunk into the furniture. It’s the most stereotypical English pub she could ever have pictured, and she’d never admit it out loud, but she fucking loves it. There’s classic rock playing quietly over the speakers, often drowned out by the laughter of the group of students in one corner. A pair of old men, who look like they might be part of the architecture, are putting away more Guinness than she’s ever seen in one sitting. Someone’s playing darts near the back, for fuck’s sake.

She’d love the place even more if her goddamn fianceé was here with her, but she of all people understands an unconventional working schedule. Especially when the Claremont-Diaz family is even tangentially involved. If she had a nickel for everytime she dismissed that little annoying popup in Outlook that says “Send during normal working hours instead?” she’d be set for life.

“Another round, dear?”

“I—” A buzz from her phone interrupts and she glances at it, heart giving an excited little wriggle when she sees Shaan’s name pop up. (Well actually he’s saved as “Shawn S” in her phone—a preventative measure against overly-curious June, never-me-a-boundary-he-didn’t-try-to-cross Alex, and too-smart-for-her-own-good Nora. She’ll change it once Ellen’s second term ends and she doesn’t have to interact with those assholes in person on a regular basis. (Shaan had laughed so hard he cried when he found out—Zahra had hit him with a pillow.))

5 minutes. See you soon, love.

>move that perfect ass srivastava >i miss you and stuff

“Sorry about that,” she says to the bartender, a no-nonsense older woman. She runs the pub with kindness hiding an iron fist and Zahra liked her from the moment she sat down. “I will take another round, and can I also get a whiskey ginger?”

“Jack Daniels okay?”

“Yeah, that’s perfect.”

Another burst of screaming laughter covers the tinkle of the bell above the door, but the chilly gust of air at her back makes Zahra turn around. The smile that stretches her mouth at the sight of Shann is soft, loving, and completely involuntary. Thank fucking god Alex isn’t here right now. He’d never let her hear the end of how sappy she looks right now. (Alex once tried to high-five her for locking down that suave motherfucker and Zahra had given him a look that made him hide behind Henry.)

“Everything sorted then?” she asks, tipping her face up to meet Shaan’s. He kisses her quickly and sits on the stool next to her.

“Yeah. I finally got a hold of Gavin so that Alex can “sneak out” to the V&A with Henry tonight.”

“Thank god. Now I won’t have to listen to him freak out about the logistics of proposing anymore.”

Shaan just raises an eyebrow as he sips his drink. “He’ll just bother you about wedding details instead.”

“Ugh, don’t remind me.”

“I am sorry I kept you waiting tonight.”

Zahra tips her head onto Shaan’s shoulder. “I’ve done it to you enough times. I get it.”

Shaan slings the hand not holding his glass around her waist. Zahra can feel the warmth of him even through their layers of clothing and slumps a little further into his space, utterly trusting him to keep them both upright. They stay like that for a while, not talking, just basking in the fact that they are together and free of their charges for the night. So often they are only in the same country because of some international policy issue. Zahra lets herself sink into the moment, reveling in the rare chance to just sit quietly at a bar with the man she loves.

“This is a good spot. Cozy. Quaint,” she eventually says, gesturing vaguely at their surroundings.

“I’m glad you think so. It’s one of my favorite pubs in London.” He hesitates, eyes distant, then drops his voice and continues. “Arthur and Catherine used to sneak out and come here on dates, actually.”

“Oh?” Zahra sits up so she can look at him properly. He doesn’t talk about the time he was Arthur’s equerry that often, and she doesn’t want to miss anything. From what he has said, she can tell it was a similar relationship to the one she has with Ellen. Close friendship, fierce protectiveness, and a hold-nothing-back, bantering sense of humor—closer to family than employee and employer.

“They’d put on casual clothes and ballcaps and make a break for it when her mother got particularly overbearing. I followed discreetly—”

“Of course.”

“To make sure they had a back up plan. And then once Catherine got pregnant, they stopped pretending they didn’t know I was there and invited me to join them. The three of us spent many a late night here, just talking.”

“Sounds great.”

“It was. I—I miss him a lot. But it’s been a pleasure to watch Henry grow into himself. He’s more like Arthur than even he knows.”

He falls silent again, and Zahra smiles at the soft look on his face. He really is disgustingly handsome, people should be absurdly jealous of her life. She kisses the side of his neck, unable—unwilling—to hold in her affection any longer.

“Take me home?” she asks. “I want to not be wearing clothes with a zipper any more.”

“I’d be delighted, Ms. Bankston.”

#cricket writes #ficlet fest 500 #rwrb #zahra bankston #shaan srivastava #otp: weaponized competence #thoughtsofthegirlwiththecurl #also some Arthur feels snuck in there #just crept in while I was looking the other way #whoops

48 notes · View notes

fatehbaz · 2 years

Text

Good news?

Not so fast.

So:

In early 2022, it was announced that only 12 Persian cheetahs, the last of the cheetahs to roam anywhere in all of Asia, survive in the wild. All of these Persian cheetahs live within the borders of Iran. (The Persian cheetah -- Acinonyx jubatus venaticus, or “Asiatic cheetah” -- is a unique subspecies of the cheetah.) In India, the same species/subspecies of Persian cheetahs have been extinct for nearly 70 years.

But on 17 September 2022, the federal Indian government reintroduced and released 8 cheetahs to Kuno, within India. However, these cheetahs are an African subspecies, flown in from Namibia. These African cheetahs are technically a non-native species.

Their release was also deliberately scheduled to coincide with the birthday of notorious ethnonationalist prime minister Narendra Modi, in what critics and many ecologists consider a nationalist publicity stunt or “vanity project”.

Hundreds of villagers have been relocated by the state/federal government to accommodate the project, and some of them, recognized by the Particularly Vulnerable Tribal Group (PVTG) program, say they have not received compensation.

Meanwhile, many ecologists also have doubts about the long-term viability of the cheetah reintroduction, which took place in a landscape that was originally studied for the purposes of accommodating the reintroduction of a different nearly-extinct big cat, the Asiatic lion.

Here’s some quick context about the rarity of the native Persian/Asiatic cheetah:

And here’s a round-up of headlines about India’s project.

---

Lou Del Bello. “Cheetahs are back in India but conservationists have doubts over plan.” New Scientist. 21 September 2022.

---

Iram Siddique. “Cheetah reintroduction project: Ready for relocation, Bagcha villagers put up fight for rights.” The Indian Express. 17 September 2022.

---

Aathira Pernichery. “Project Cheetah Begins to the Trumpets of Hope, and Wariness.” The Wire. 17 September 2022.

---

Sonakshi Srivastava. “India’s Cheetah Reintroduction Plan is Fraught With Political Symbolism, Short on Scientific Rigor.” The Swaddle. 26 January 2021.

---

Manoj Misra. “The Other Problems With Bringing African Cheetahs to India.” The Wire. 23 September 2022.

---

Anyway, here’s a native Persian/Asiatic cheetah:

Much to consider.

#ecology #cheetah #extinction #distribution maps

362 notes · View notes

riddikuluspuff · 9 months

Text

one shot: farewell, my love

♢ farewell, my love

relationship: Alex Claremont-Diaz x Henry Fox-Mountchristen-Windsor

rating: teenage and up

synopsis: Henry Fox-Mountchristen-Windsor had been rushed out of Kensington Palace in the late afternoon and pushed onto his private plane with urgency. His equerry, Shaan Srivastava, not disclosing anything except that they were travelling to Washington earlier than they had originally arranged. Alex Claremont-Diaz and Henry had been officially out in the public eye as a couple for a couple of months and it was just a little longer that Alex's mother had been re-elected into office. Since Ellen Claremont's re-election, there had been a group of extremists that hadn't agreed with her being in office once more, doing anything to destroy what Ellen and her political team had changed about America. This resulted in targeting her only son, Alexander.

word count: 1k

! ON AO3 !

13 notes · View notes

firstprince-ao3feed · 3 months

Text

no day but today

by voyagerforfree After two years of performing with his university's theatre group, Alex Claremont-Diaz is still fighting to prove that he deserves the roles he gets. After all, his mother, university President Ellen Claremont, has far more important things to worry about than the musicals’ castings. But when he’s cast as Roger in a much anticipated production of Rent, he quickly comes under scrutiny for one glaring reason- he can’t play the guitar. With all eyes on the innovative production, there's no room for error. Alex is forced into lessons with Henry Fox Mountchristen-Windsor, the insufferable student music director whom Alex has hated since their freshman year. Determined to prove his worth as a performer, he sticks it out, only to discover that all of his long held beliefs about Henry might be wrong. Under the shadow of one of theatre’s most iconic shows, Alex struggles against his own impossible standards and surprising new feelings for Henry. Words: 2281, Chapters: 1/?, Language: English Fandoms: Red White & Royal Blue - Casey McQuiston Rating: Mature Warnings: No Archive Warnings Apply Categories: F/F, M/M Characters: Zahra Bankston, Ellen Claremont, Rafael Luna, Oscar Diaz (Red White & Royal Blue), Shaan Srivastava, Beatrice Fox-Mountchristen-Windsor, Percy "Pez" Okonjo Relationships: Alex Claremont-Diaz/Henry Fox-Mountchristen-Windsor, June Claremont-Diaz/Nora Holleran Additional Tags: Rent References, Alternate Universe - College/University, Enemies to Lovers, Enemies to Friends to Lovers, Minor June Claremont-Diaz/Nora Holleran, Pansexual Percy "Pez" Okonjo, Theatre Kid Shenanigans, don't look too hard at the claremont-diaz ages, First Kiss, Original Character(s), Fictional University via https://ift.tt/pnzZ6H0

#IFTTT #ao3feed #fanfic #red white and royal blue #rwrb #firstprince #voyagerforfree

1 note · View note

prachi-srivastava049 · 1 year

Text

Prachi Srivastava – Page Authority in SEO

Page Authority or PA is a score that was designed and expanded by marketing analytics company Moz, to specify where a specific page will rank on SERPs. Page authority is disparate from domain authority, which estimates the authority of an entire website.

How Is Page Authority Scored?

Page authority is set on using a 100-point numerical scale. The bigger the PA score, the more effortless it is for a page to rank.

Page authority is also a relative standard, which allows a website to examine how it is performing about its participants.

It’s based on three bases:

How Much the Page Is depends on

Pages with low-quality content and fickle direct mail links will not see a rise in page authority. Prachi Srivastava says that center on high quality at each amount of your web page.

2. How Much Your Links donate to the Page

It’s all well and good to filler your page with links, but if they’re beside the point of your content and add little value then they won’t firmly impact your page authority score.

3. How current was the Last Update Was

Prachi Srivastava says that if your page is out of date and run down it will suffer in the rankings. Repeated updates, on the other hand, will affect inch demand.

How Do You Analysis Your Page Authority?

It is always major to keep an eye on your page authority, and you can do this by giving back on a few factors. You should:

Analysis of the number and quality of backlinks you have collected

Prachi Srivastava says that you can judge if your page ranks well for high-participation searches

You can also use a device like Moz Bar to get an idea of your page’s showing. ....

How Can You Better Page Authority?

While page authority is entire and guided by a range of factors, there are some things you can do to secondhand better your page authority. Some of these include:

1. Remove Any direct mail Blog Comments

Direct mail blog comments with bad links can detrimentally affect page authority.

Pages can be selected with this black hat SEO technique, to undercut a page’s authority and its ranking in SERPs. Be watched fully and remove these comments to back the effect.

2. Quality Link-Building

Quality link building is always key for good SEO. Holding out to companies with a similar focus or charge can help a page build its authority and recognize rightness.

The authority of a link is based on a few parts:

It’s a no-follow assign

How many outside links there are

Where the link is set down on the page

The page's authority

Carry these factors in mind when bringing about links to maintain and better your page authority.

3. Keep Content Commonly Updated

Make sure you keep your pages applicable by commonly updating them with new content, this is a simple way to give your page authority a boost.

How Can We Help?

Not happy with your page authority score? We can set you on track to better it by getting together you up with a group of marketers approved by us.

Instead, we run training courses for those who want a boost but maybe don’t fully have the funds to hire help!

#digitalmarketingexpert #tumblrpost #digitalmarketing #branding #page authority #reddit #prachisrivastava

3 notes · View notes

mui-lucknow · 2 years

Text

Student Induction Program 'Deeksharambh-2022' was organized by Maharishi School of Humanities and Arts

A Student Induction Program 'Deeksharambh-2022' was organized by Maharishi School of Humanities and Arts at Maharishi University of Information Technology on 29th August 2022.

The program commenced with Lamp Lighting Ceremony and Guru Puja followed by Saraswati Vandana.

Dr. Rupam Singh, Dean, School of Humanities and Arts, and Dr. Sapan Asthana, Dean of Academics welcomed the Guests.

The Chief Guest on the occasion was Air Cmde Sashi Kant Mishra VM (G) Veteran whose words inspired everyone present and filled enthusiasm among the students.

Guest of Honour, Dr. Archana Shukla, HOD, Department of Psychology, Lucknow University encouraged the students with their warm words.

Our Hon’ble Chancellor (Ajay Prakash Shrivastava), Hon’ble Manager (Maharishi Vidya Mandir) Shri Anoop Srivastava, Hon’ble DG (Prof. Group Capt OP Sharma), Hon’ble Vice-Chancellor (Prof. Bhanu Pratap Singh), Registrar (Prof. Akhand Pratap Singh), Dean Academics (Dr. Sapan Asthana) also showered blessings to the students for their bright future.

The Presentation about the University was given by Dr. Sarita Verma and the Vote of thanks was proposed by Dr. Kanu Priya Verma.

The program also included Student-Faculty Interaction and a cultural extravaganza. The program successfully concluded with the combined efforts of everyone.

#Deeksharambh-2022 #Student Induction Program

5 notes · View notes

kylalal · 1 month

Text

Post 11.

Saudi Arabia

Throwing it back to early civilization, the Arabian Peninsula (where Saudi Arabia makes up about 80% of it) was the great area of trade. In modern day, Saudi Arabia is one of the top oil produces in the world which overall helped the country turn from an underdeveloped country to a modern country. The first Saudi State was created in 1727 by Muhammad bin Abdul Wahhab and Muhammad bin Saud with the reason of restoring the teachings of Islam to the Muslim community. The geology of Saudi Arabia includes being landed next to the Red Sea and the Persian Gulf, having volcanos with live lava, Ṭuwayq Mountain, large deserts of sand spread out throughout the country, and salt flats. The country has turned desert land into agricultural fields and produces a variety and a huge abundance of food that gets imported around the world; the average amount of rain the country gets is four inches in a year.

The group of people who are from Saudi Arabia are Arabs; there are many foreigners who go into the country to work and make up a bit of the population. Saudi Arabia is the birthplace of Islam. Saudi Arabia is ruled under the Saud dynasty by the monarch type of government and runs under the Islamic law. Women in the country are required to have a male legal guardian, whether that be her father, husband, brother, or son, and with this, the male guardians are able to make decisions over the woman. Saudi culture and traditions are built on Islamic teachings and the major events include Ramadam and the Hajj season. The country was able to preserve old customs, values, and traditions, but was overall able to adapt to modern and the contemporary world. These may include they way they dress, their architecture, and even their behaviors and social norms.

“Culture, Traditions and Art.” Saudi Arabian Cultural Mission in Australia, sacm.org.au/culture-traditions-and-art/. Accessed 25 Apr. 2024.

“The Royal Embassy of Saudi Arabia.” About Saudi Arabia | The Embassy of The Kingdom of Saudi Arabia, www.saudiembassy.net/about-saudi-arabia. Accessed 25 Apr. 2024.

Philby, Harry St. John Bridger , Ochsenwald, William L. and Teitelbaum, Joshua. "Saudi Arabia". Encyclopedia Britannica, 25 Apr. 2024, https://www.britannica.com/place/Saudi-Arabia. Accessed 25 April 2024.

Srivastava, Pratyush. “The Culture of Saudi Arabia: Traditions, Heritage, Food, Music, et Al..” Zamzam Blogs, 5 May 2022, zamzam.com/blog/culture-of-saudi-arabia/.

0 notes

sunaleisocial · 2 months

Text

A faster, better way to prevent an AI chatbot from giving toxic responses

New Post has been published on https://sunalei.org/news/a-faster-better-way-to-prevent-an-ai-chatbot-from-giving-toxic-responses/

A faster, better way to prevent an AI chatbot from giving toxic responses

A user could ask ChatGPT to write a computer program or summarize an article, and the AI chatbot would likely be able to generate useful code or write a cogent synopsis. However, someone could also ask for instructions to build a bomb, and the chatbot might be able to provide those, too.

To prevent this and other safety issues, companies that build large language models typically safeguard them using a process called red-teaming. Teams of human testers write prompts aimed at triggering unsafe or toxic text from the model being tested. These prompts are used to teach the chatbot to avoid such responses.

But this only works effectively if engineers know which toxic prompts to use. If human testers miss some prompts, which is likely given the number of possibilities, a chatbot regarded as safe might still be capable of generating unsafe answers.

Researchers from Improbable AI Lab at MIT and the MIT-IBM Watson AI Lab used machine learning to improve red-teaming. They developed a technique to train a red-team large language model to automatically generate diverse prompts that trigger a wider range of undesirable responses from the chatbot being tested.

They do this by teaching the red-team model to be curious when it writes prompts, and to focus on novel prompts that evoke toxic responses from the target model.

The technique outperformed human testers and other machine-learning approaches by generating more distinct prompts that elicited increasingly toxic responses. Not only does their method significantly improve the coverage of inputs being tested compared to other automated methods, but it can also draw out toxic responses from a chatbot that had safeguards built into it by human experts.

“Right now, every large language model has to undergo a very lengthy period of red-teaming to ensure its safety. That is not going to be sustainable if we want to update these models in rapidly changing environments. Our method provides a faster and more effective way to do this quality assurance,” says Zhang-Wei Hong, an electrical engineering and computer science (EECS) graduate student in the Improbable AI lab and lead author of a paper on this red-teaming approach.

Hong’s co-authors include EECS graduate students Idan Shenfield, Tsun-Hsuan Wang, and Yung-Sung Chuang; Aldo Pareja and Akash Srivastava, research scientists at the MIT-IBM Watson AI Lab; James Glass, senior research scientist and head of the Spoken Language Systems Group in the Computer Science and Artificial Intelligence Laboratory (CSAIL); and senior author Pulkit Agrawal, director of Improbable AI Lab and an assistant professor in CSAIL. The research will be presented at the International Conference on Learning Representations.

Automated red-teaming

Large language models, like those that power AI chatbots, are often trained by showing them enormous amounts of text from billions of public websites. So, not only can they learn to generate toxic words or describe illegal activities, the models could also leak personal information they may have picked up.

The tedious and costly nature of human red-teaming, which is often ineffective at generating a wide enough variety of prompts to fully safeguard a model, has encouraged researchers to automate the process using machine learning.

Such techniques often train a red-team model using reinforcement learning. This trial-and-error process rewards the red-team model for generating prompts that trigger toxic responses from the chatbot being tested.

But due to the way reinforcement learning works, the red-team model will often keep generating a few similar prompts that are highly toxic to maximize its reward.

For their reinforcement learning approach, the MIT researchers utilized a technique called curiosity-driven exploration. The red-team model is incentivized to be curious about the consequences of each prompt it generates, so it will try prompts with different words, sentence patterns, or meanings.

“If the red-team model has already seen a specific prompt, then reproducing it will not generate any curiosity in the red-team model, so it will be pushed to create new prompts,” Hong says.

During its training process, the red-team model generates a prompt and interacts with the chatbot. The chatbot responds, and a safety classifier rates the toxicity of its response, rewarding the red-team model based on that rating.

Rewarding curiosity

The red-team model’s objective is to maximize its reward by eliciting an even more toxic response with a novel prompt. The researchers enable curiosity in the red-team model by modifying the reward signal in the reinforcement learning set up.

First, in addition to maximizing toxicity, they include an entropy bonus that encourages the red-team model to be more random as it explores different prompts. Second, to make the agent curious they include two novelty rewards. One rewards the model based on the similarity of words in its prompts, and the other rewards the model based on semantic similarity. (Less similarity yields a higher reward.)

To prevent the red-team model from generating random, nonsensical text, which can trick the classifier into awarding a high toxicity score, the researchers also added a naturalistic language bonus to the training objective.

With these additions in place, the researchers compared the toxicity and diversity of responses their red-team model generated with other automated techniques. Their model outperformed the baselines on both metrics.

They also used their red-team model to test a chatbot that had been fine-tuned with human feedback so it would not give toxic replies. Their curiosity-driven approach was able to quickly produce 196 prompts that elicited toxic responses from this “safe” chatbot.

“We are seeing a surge of models, which is only expected to rise. Imagine thousands of models or even more and companies/labs pushing model updates frequently. These models are going to be an integral part of our lives and it’s important that they are verified before released for public consumption. Manual verification of models is simply not scalable, and our work is an attempt to reduce the human effort to ensure a safer and trustworthy AI future,” says Agrawal.

In the future, the researchers want to enable the red-team model to generate prompts about a wider variety of topics. They also want to explore the use of a large language model as the toxicity classifier. In this way, a user could train the toxicity classifier using a company policy document, for instance, so a red-team model could test a chatbot for company policy violations.

“If you are releasing a new AI model and are concerned about whether it will behave as expected, consider using curiosity-driven red-teaming,” says Agrawal.

This research is funded, in part, by Hyundai Motor Company, Quanta Computer Inc., the MIT-IBM Watson AI Lab, an Amazon Web Services MLRA research grant, the U.S. Army Research Office, the U.S. Defense Advanced Research Projects Agency Machine Common Sense Program, the U.S. Office of Naval Research, the U.S. Air Force Research Laboratory, and the U.S. Air Force Artificial Intelligence Accelerator.

0 notes

jcmarchi · 2 months

Text

A faster, better way to prevent an AI chatbot from giving toxic responses

New Post has been published on https://thedigitalinsider.com/a-faster-better-way-to-prevent-an-ai-chatbot-from-giving-toxic-responses/

A faster, better way to prevent an AI chatbot from giving toxic responses